diff --git a/.gitignore b/.gitignore
index c2bbe26..1e0b755 100644
--- a/.gitignore
+++ b/.gitignore
@@ -4,6 +4,7 @@
 # runtime state, regenerated by the server, not source
 live-state.json
 dance-state.json
+library.json
 
 # scratch server logs, if any get written locally
 *.log
diff --git a/library.json b/library.json
deleted file mode 100644
index 8fb8b4f..0000000
--- a/library.json
+++ /dev/null
@@ -1,5247 +0,0 @@
-[
-  {
-    "id": "20332b26881c",
-    "source": "h3console",
-    "prompt": "For the target video, at 0.00 seconds into the target video, (from [Shot 1]) is fully referenced. integrated_multimodal_",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nintegrated_multimodal_description:\n[Shot 1] The shot begins from <Picture 1>: a live-action mid-2000s office mockumentary frame inside a fluorescent paper-company office. A male regional manager in a dark suit and patterned tie crosses the foreground near the reception desk. A female receptionist sits behind the counter, with office cubicles, stacked paper boxes, a wall calendar, framed certificates, and a standing lamp visible around them. Preserve the handheld documentary camera feel, muted beige office colors, natural fluorescent lighting, and awkward workplace realism. No subtitles, no captions, no UI overlays, no watermark, and no newly generated on-screen text.\n[Shot 2] At 00:01.200, as the manager gestures through the office tour, a tiny orange flame suddenly flickers from a small stack of loose papers near the reception counter, beside the paper boxes. The flame is small and localized, more embarrassing than dangerous. The camera gives a quick handheld wobble toward the flame. The receptionist notices first and stiffens. The manager tries to keep his tour face and says: <d>[English] This is not a fire. This is urgency.</d>\n[Shot 3] At 00:03.400, the receptionist calmly reaches for a coffee mug or small cup of water on the desk. The manager steps half a pace toward the camera, trying to block the view while still smiling awkwardly. The flame remains small, with a faint curl of smoke. Keep the action realistic, contained, and office-scale. The receptionist says softly: <d>[English] Michael.</d>\n[Shot 4] At 00:05.200, cut to a dry mockumentary reaction angle from a nearby salesman at his desk in the bullpen. He glances at the tiny flame, then looks into the camera with a deadpan expression. The salesman says: <d>[English] Technically, paper is the business.</d> The background stays consistent with the same office.\n[Shot 5] At 00:06.700, cut back near reception. An intense coworker rushes in from the side holding a fire extinguisher too seriously, but the receptionist has already poured a small splash of water onto the papers. The flame disappears with a little hiss and a modest puff of smoke. The intense coworker freezes, disappointed that the emergency is over. The manager says: <d>[English] Handled. That's management.</d>\n[Shot 6] At 00:09.000, the receptionist quietly moves the damp papers aside while the manager smooths his suit jacket and tries to restart the office tour as if nothing happened. The intense coworker lowers the extinguisher with excessive seriousness. The camera settles back into the reception-tour composition, matching <Picture 1> as closely as possible: same office angle, manager in the foreground near reception, receptionist behind the counter, cubicles and paper stacks visible.\n[Shot 7] At 00:11.300, use a tiny documentary-style jump cut to a near match of <Picture 1>. No visible flame remains, only normal office room tone and a barely noticeable wisp of smoke. By 00:12.250, the frame should be ready for the source episode to resume from the same reception-tour moment.\n\noverall_soundscape:\nNatural office room tone throughout: fluorescent hum, quiet desk movement, paper rustle, a small flame crackle, a quick splash of water, a faint hiss, and restrained workplace reactions. Dialogue is spoken in English only and must never appear as subtitles or captions.\n\nnon_diegetic_music:\nN/A",
-    "edited": false,
-    "width": 832,
-    "height": 480,
-    "duration": 12.3,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788330575546
-  },
-  {
-    "id": "3d825658c100",
-    "source": "h3console",
-    "edited": true,
-    "channel": "dance",
-    "characterId": "4a4dc93b-4984-495b-a660-535ce3d3fc51",
-    "garmentId": "9a9dd316-1178-4b10-801f-64a0572e82bc",
-    "refDanceVideoId": "7b8e13ce-c9e3-4a19-bccc-5f54e6aae98e",
-    "backend": "h3_console",
-    "priority": "normal",
-    "requestedByUserId": null,
-    "prompt": "验证-人物 × 紫色旗袍",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the young woman whose visual identity is defined by <Picture 1>: 年轻女性，长直黑发，鹅蛋脸，气质清冷. Preserve her face, hairstyle, and body exactly as shown; only her outfit, pose, motion, and framing change in the target video.\n<Picture 1> is the character reference frame for <Subject 1>, showing her face, hairstyle, and body in a neutral standing pose, wearing a neutral outfit that does not appear in the target video; her footwear as shown here is retained unchanged in the target video.\n<Picture 2> is the garment reference image: 紫色旗袍，浅粉滚边，盘扣，侧开衩, laid flat with no person wearing it. Only its garment attributes (silhouette, color, fabric, trim, cut) transfer onto <Subject 1>; no other content of this image appears in the target video, and it does not affect her footwear.\n<Video 1> is the reference choreography clip. It defines the dance structure, phrasing, and camera rhythm that the target video follows; it is not a visual identity source, and none of its performer, background, or imagery appears in the target video.\n<Audio 1> is the background music track carried on <Video 1>, reused verbatim as the music bed of the target video. It contains no speech.\n\nsummary:\n[reference generation + attribute transfer + audio reuse] The target video shows <Subject 1>, now wearing the outfit transferred from <Picture 2>, performing 宽马步，双臂交替甩动，配合甩发动作，节奏感强的hip-hop风格 in a vertical studio setting, following the pacing and camera rhythm established in <Video 1>, with the music of <Audio 1> reused as its track.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - her face, hairstyle, and body are retained unchanged throughout; her outfit changes to the garment shown in <Picture 2>, her footwear stays as shown in <Picture 1>.\n<Picture 1>: fully_preserved - it supplies the face, hairstyle, body, and footwear of <Subject 1> and the opening framing of the shot.\n<Picture 2>: attribute_transfer - only the garment's silhouette, color, fabric, and trim transfer onto <Subject 1>; the image itself does not appear.\n<Video 1>: weak_reference - only the choreography phrasing and the camera rhythm are referenced; no performer, background, or imagery from it is retained.\n<Audio 1>: fully_copy - the music track is reused verbatim as the non-diegetic music bed.\n\ndetailed_description:\nA vibrant, mobile-first vertical dance video with warm, glossy studio lighting, saturated color, and a clean contemporary look sustained from the first frame to the last. [Shot 1] The shot begins from <Picture 1>, with <Subject 1> now wearing the outfit transferred from <Picture 2>. She stands centered in a medium-full vertical frame against a softly lit seamless backdrop that falls off into gentle shadow toward the edges. A warm key light sits slightly camera-left, a cooler rim light separates her shoulders and hair from the background, and a soft fill keeps her face evenly exposed. She begins in a relaxed neutral stance, weight settled on her back foot, hands loose at her sides, chin level with the lens. On the first musical accent she starts 宽马步，双臂交替甩动，配合甩发动作，节奏感强的hip-hop风格, and the routine unfolds continuously from there with no cut. Her weight transfers smoothly from foot to foot on the beat, hips leading each transfer and shoulders following a fraction behind, so the movement reads as one connected chain rather than a series of separate poses. Her arms trace clean, unhurried arcs that extend fully at the peak of each phrase and soften on the release, wrists relaxed and fingers naturally extended. Between phrases she adds small directional changes: a quarter turn away from the lens and back, a step that closes distance with the camera, a shift of gaze that returns to the lens on the downbeat. The camera holds a Static Shot through the opening beats, then performs a slow Push In with small amplitude at slow speed, tightening from medium-full to medium while keeping her head and hands inside the frame at every point of the routine. Her expression stays confident and playful, with a light smile and eyes that meet the lens at the end of each phrase. The fabric of the transferred outfit swings and settles naturally with each step, turn, and arm extension, and her hair carries the momentum of her head movement half a beat behind it. The pacing, phrasing, and camera rhythm follow the structure established in <Video 1>: the routine builds energy through the middle of the clip, reaches its widest and fastest movement around the two-thirds point, then settles back toward stillness. The closing beats return her to a stance, framing, and lighting that closely match her opening position, so the clip loops cleanly for repeated live playback. No text, graphics, watermarks, or overlays appear anywhere in the frame at any point.\n\noverall_soundscape:\nSoft ambient room tone of an enclosed studio sits under the music, with faint fabric rustle as <Subject 1> moves and light rhythmic footfall on the floor as her weight transfers. No speech, no crowd, and no off-screen sound events occur at any point.\n\nnon_diegetic_music:\nAn upbeat instrumental track drives the routine from the first frame to the last; its tempo, instrumentation, and dynamics are copied verbatim from <Audio 1>. Its accents align with the weight transfers and arm extensions of <Subject 1>, and its energy build through the middle of the clip matches the build in her movement.",
-    "width": 480,
-    "height": 832,
-    "duration": 10,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788330297866,
-    "lastPlayedAt": 1788331519272,
-    "playCount": 46
-  },
-  {
-    "id": "e7f500b05262",
-    "source": "h3console",
-    "edited": true,
-    "channel": "dance",
-    "characterId": "4a4dc93b-4984-495b-a660-535ce3d3fc51",
-    "garmentId": "69373e07-2abf-4b26-819a-fcae2127e087",
-    "refDanceVideoId": "7b8e13ce-c9e3-4a19-bccc-5f54e6aae98e",
-    "backend": "h3_console",
-    "priority": "normal",
-    "requestedByUserId": null,
-    "prompt": "验证-人物 × 黄绿色旗袍",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the young woman whose visual identity is defined by <Picture 1>: 年轻女性，长直黑发，鹅蛋脸，气质清冷. Preserve her face, hairstyle, and body exactly as shown; only her outfit, pose, motion, and framing change in the target video.\n<Picture 1> is the character reference frame for <Subject 1>, showing her face, hairstyle, and body in a neutral standing pose, wearing a neutral outfit that does not appear in the target video; her footwear as shown here is retained unchanged in the target video.\n<Picture 2> is the garment reference image: 黄绿色旗袍，紫色滚边，盘扣，侧开衩, laid flat with no person wearing it. Only its garment attributes (silhouette, color, fabric, trim, cut) transfer onto <Subject 1>; no other content of this image appears in the target video, and it does not affect her footwear.\n<Video 1> is the reference choreography clip. It defines the dance structure, phrasing, and camera rhythm that the target video follows; it is not a visual identity source, and none of its performer, background, or imagery appears in the target video.\n<Audio 1> is the background music track carried on <Video 1>, reused verbatim as the music bed of the target video. It contains no speech.\n\nsummary:\n[reference generation + attribute transfer + audio reuse] The target video shows <Subject 1>, now wearing the outfit transferred from <Picture 2>, performing 宽马步，双臂交替甩动，配合甩发动作，节奏感强的hip-hop风格 in a vertical studio setting, following the pacing and camera rhythm established in <Video 1>, with the music of <Audio 1> reused as its track.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - her face, hairstyle, and body are retained unchanged throughout; her outfit changes to the garment shown in <Picture 2>, her footwear stays as shown in <Picture 1>.\n<Picture 1>: fully_preserved - it supplies the face, hairstyle, body, and footwear of <Subject 1> and the opening framing of the shot.\n<Picture 2>: attribute_transfer - only the garment's silhouette, color, fabric, and trim transfer onto <Subject 1>; the image itself does not appear.\n<Video 1>: weak_reference - only the choreography phrasing and the camera rhythm are referenced; no performer, background, or imagery from it is retained.\n<Audio 1>: fully_copy - the music track is reused verbatim as the non-diegetic music bed.\n\ndetailed_description:\nA vibrant, mobile-first vertical dance video with warm, glossy studio lighting, saturated color, and a clean contemporary look sustained from the first frame to the last. [Shot 1] The shot begins from <Picture 1>, with <Subject 1> now wearing the outfit transferred from <Picture 2>. She stands centered in a medium-full vertical frame against a softly lit seamless backdrop that falls off into gentle shadow toward the edges. A warm key light sits slightly camera-left, a cooler rim light separates her shoulders and hair from the background, and a soft fill keeps her face evenly exposed. She begins in a relaxed neutral stance, weight settled on her back foot, hands loose at her sides, chin level with the lens. On the first musical accent she starts 宽马步，双臂交替甩动，配合甩发动作，节奏感强的hip-hop风格, and the routine unfolds continuously from there with no cut. Her weight transfers smoothly from foot to foot on the beat, hips leading each transfer and shoulders following a fraction behind, so the movement reads as one connected chain rather than a series of separate poses. Her arms trace clean, unhurried arcs that extend fully at the peak of each phrase and soften on the release, wrists relaxed and fingers naturally extended. Between phrases she adds small directional changes: a quarter turn away from the lens and back, a step that closes distance with the camera, a shift of gaze that returns to the lens on the downbeat. The camera holds a Static Shot through the opening beats, then performs a slow Push In with small amplitude at slow speed, tightening from medium-full to medium while keeping her head and hands inside the frame at every point of the routine. Her expression stays confident and playful, with a light smile and eyes that meet the lens at the end of each phrase. The fabric of the transferred outfit swings and settles naturally with each step, turn, and arm extension, and her hair carries the momentum of her head movement half a beat behind it. The pacing, phrasing, and camera rhythm follow the structure established in <Video 1>: the routine builds energy through the middle of the clip, reaches its widest and fastest movement around the two-thirds point, then settles back toward stillness. The closing beats return her to a stance, framing, and lighting that closely match her opening position, so the clip loops cleanly for repeated live playback. No text, graphics, watermarks, or overlays appear anywhere in the frame at any point.\n\noverall_soundscape:\nSoft ambient room tone of an enclosed studio sits under the music, with faint fabric rustle as <Subject 1> moves and light rhythmic footfall on the floor as her weight transfers. No speech, no crowd, and no off-screen sound events occur at any point.\n\nnon_diegetic_music:\nAn upbeat instrumental track drives the routine from the first frame to the last; its tempo, instrumentation, and dynamics are copied verbatim from <Audio 1>. Its accents align with the weight transfers and arm extensions of <Subject 1>, and its energy build through the middle of the clip matches the build in her movement.",
-    "width": 480,
-    "height": 832,
-    "duration": 10,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788329156129,
-    "lastPlayedAt": 1788331509252,
-    "playCount": 109
-  },
-  {
-    "id": "cfefb1114e41",
-    "source": "h3console",
-    "edited": true,
-    "channel": "dance",
-    "characterId": "4a4dc93b-4984-495b-a660-535ce3d3fc51",
-    "garmentId": "a536decb-eeef-4a3e-88d7-ad45c39ccb23",
-    "refDanceVideoId": "7b8e13ce-c9e3-4a19-bccc-5f54e6aae98e",
-    "backend": "h3_console",
-    "priority": "normal",
-    "requestedByUserId": null,
-    "prompt": "验证-人物 × 绿色旗袍",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the young woman whose visual identity is defined by <Picture 1>: 年轻女性，长直黑发，鹅蛋脸，气质清冷. Preserve her face, hairstyle, and body exactly as shown; only her outfit, pose, motion, and framing change in the target video.\n<Picture 1> is the character reference frame for <Subject 1>, showing her face, hairstyle, and body in a neutral standing pose, wearing a neutral outfit that does not appear in the target video; her footwear as shown here is retained unchanged in the target video.\n<Picture 2> is the garment reference image: 绿色旗袍，粉色滚边，盘扣，侧开衩, laid flat with no person wearing it. Only its garment attributes (silhouette, color, fabric, trim, cut) transfer onto <Subject 1>; no other content of this image appears in the target video, and it does not affect her footwear.\n<Video 1> is the reference choreography clip. It defines the dance structure, phrasing, and camera rhythm that the target video follows; it is not a visual identity source, and none of its performer, background, or imagery appears in the target video.\n<Audio 1> is the background music track carried on <Video 1>, reused verbatim as the music bed of the target video. It contains no speech.\n\nsummary:\n[reference generation + attribute transfer + audio reuse] The target video shows <Subject 1>, now wearing the outfit transferred from <Picture 2>, performing 宽马步，双臂交替甩动，配合甩发动作，节奏感强的hip-hop风格 in a vertical studio setting, following the pacing and camera rhythm established in <Video 1>, with the music of <Audio 1> reused as its track.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - her face, hairstyle, and body are retained unchanged throughout; her outfit changes to the garment shown in <Picture 2>, her footwear stays as shown in <Picture 1>.\n<Picture 1>: fully_preserved - it supplies the face, hairstyle, body, and footwear of <Subject 1> and the opening framing of the shot.\n<Picture 2>: attribute_transfer - only the garment's silhouette, color, fabric, and trim transfer onto <Subject 1>; the image itself does not appear.\n<Video 1>: weak_reference - only the choreography phrasing and the camera rhythm are referenced; no performer, background, or imagery from it is retained.\n<Audio 1>: fully_copy - the music track is reused verbatim as the non-diegetic music bed.\n\ndetailed_description:\nA vibrant, mobile-first vertical dance video with warm, glossy studio lighting, saturated color, and a clean contemporary look sustained from the first frame to the last. [Shot 1] The shot begins from <Picture 1>, with <Subject 1> now wearing the outfit transferred from <Picture 2>. She stands centered in a medium-full vertical frame against a softly lit seamless backdrop that falls off into gentle shadow toward the edges. A warm key light sits slightly camera-left, a cooler rim light separates her shoulders and hair from the background, and a soft fill keeps her face evenly exposed. She begins in a relaxed neutral stance, weight settled on her back foot, hands loose at her sides, chin level with the lens. On the first musical accent she starts 宽马步，双臂交替甩动，配合甩发动作，节奏感强的hip-hop风格, and the routine unfolds continuously from there with no cut. Her weight transfers smoothly from foot to foot on the beat, hips leading each transfer and shoulders following a fraction behind, so the movement reads as one connected chain rather than a series of separate poses. Her arms trace clean, unhurried arcs that extend fully at the peak of each phrase and soften on the release, wrists relaxed and fingers naturally extended. Between phrases she adds small directional changes: a quarter turn away from the lens and back, a step that closes distance with the camera, a shift of gaze that returns to the lens on the downbeat. The camera holds a Static Shot through the opening beats, then performs a slow Push In with small amplitude at slow speed, tightening from medium-full to medium while keeping her head and hands inside the frame at every point of the routine. Her expression stays confident and playful, with a light smile and eyes that meet the lens at the end of each phrase. The fabric of the transferred outfit swings and settles naturally with each step, turn, and arm extension, and her hair carries the momentum of her head movement half a beat behind it. The pacing, phrasing, and camera rhythm follow the structure established in <Video 1>: the routine builds energy through the middle of the clip, reaches its widest and fastest movement around the two-thirds point, then settles back toward stillness. The closing beats return her to a stance, framing, and lighting that closely match her opening position, so the clip loops cleanly for repeated live playback. No text, graphics, watermarks, or overlays appear anywhere in the frame at any point.\n\noverall_soundscape:\nSoft ambient room tone of an enclosed studio sits under the music, with faint fabric rustle as <Subject 1> moves and light rhythmic footfall on the floor as her weight transfers. No speech, no crowd, and no off-screen sound events occur at any point.\n\nnon_diegetic_music:\nAn upbeat instrumental track drives the routine from the first frame to the last; its tempo, instrumentation, and dynamics are copied verbatim from <Audio 1>. Its accents align with the weight transfers and arm extensions of <Subject 1>, and its energy build through the middle of the clip matches the build in her movement.",
-    "width": 480,
-    "height": 832,
-    "duration": 10,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788328011033,
-    "lastPlayedAt": 1788331529280,
-    "playCount": 109
-  },
-  {
-    "id": "76b6d62d832c",
-    "source": "h3console",
-    "edited": true,
-    "channel": "dance",
-    "characterId": "4a4dc93b-4984-495b-a660-535ce3d3fc51",
-    "garmentId": "504250f3-49a9-4937-822b-3dca2462b136",
-    "refDanceVideoId": "e5e56a4f-2318-4fec-91be-fd94cf5dcdfc",
-    "backend": "h3_console",
-    "priority": "normal",
-    "requestedByUserId": null,
-    "prompt": "验证-人物 × 宝蓝色旗袍",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the young woman whose visual identity is defined by <Picture 1>: 年轻女性，长直黑发，鹅蛋脸，气质清冷. Preserve her face, hairstyle, and body exactly as shown; only her outfit, pose, motion, and framing change in the target video.\n<Picture 1> is the character reference frame for <Subject 1>, showing her face, hairstyle, and body in a neutral standing pose, wearing a neutral outfit that does not appear in the target video; her footwear as shown here is retained unchanged in the target video.\n<Picture 2> is the garment reference image: 宝蓝色旗袍，淡紫滚边，盘扣，侧开衩, laid flat with no person wearing it. Only its garment attributes (silhouette, color, fabric, trim, cut) transfer onto <Subject 1>; no other content of this image appears in the target video, and it does not affect her footwear.\n<Video 1> is the reference choreography clip. It defines the dance structure, phrasing, and camera rhythm that the target video follows; it is not a visual identity source, and none of its performer, background, or imagery appears in the target video.\n<Audio 1> is the background music track carried on <Video 1>, reused verbatim as the music bed of the target video. It contains no speech.\n\nsummary:\n[reference generation + attribute transfer + audio reuse] The target video shows <Subject 1>, now wearing the outfit transferred from <Picture 2>, performing 原地小幅度手势舞动，柔和克制的比心和摆手动作，节奏偏慢 in a vertical studio setting, following the pacing and camera rhythm established in <Video 1>, with the music of <Audio 1> reused as its track.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - her face, hairstyle, and body are retained unchanged throughout; her outfit changes to the garment shown in <Picture 2>, her footwear stays as shown in <Picture 1>.\n<Picture 1>: fully_preserved - it supplies the face, hairstyle, body, and footwear of <Subject 1> and the opening framing of the shot.\n<Picture 2>: attribute_transfer - only the garment's silhouette, color, fabric, and trim transfer onto <Subject 1>; the image itself does not appear.\n<Video 1>: weak_reference - only the choreography phrasing and the camera rhythm are referenced; no performer, background, or imagery from it is retained.\n<Audio 1>: fully_copy - the music track is reused verbatim as the non-diegetic music bed.\n\ndetailed_description:\nA vibrant, mobile-first vertical dance video with warm, glossy studio lighting, saturated color, and a clean contemporary look sustained from the first frame to the last. [Shot 1] The shot begins from <Picture 1>, with <Subject 1> now wearing the outfit transferred from <Picture 2>. She stands centered in a medium-full vertical frame against a softly lit seamless backdrop that falls off into gentle shadow toward the edges. A warm key light sits slightly camera-left, a cooler rim light separates her shoulders and hair from the background, and a soft fill keeps her face evenly exposed. She begins in a relaxed neutral stance, weight settled on her back foot, hands loose at her sides, chin level with the lens. On the first musical accent she starts 原地小幅度手势舞动，柔和克制的比心和摆手动作，节奏偏慢, and the routine unfolds continuously from there with no cut. Her weight transfers smoothly from foot to foot on the beat, hips leading each transfer and shoulders following a fraction behind, so the movement reads as one connected chain rather than a series of separate poses. Her arms trace clean, unhurried arcs that extend fully at the peak of each phrase and soften on the release, wrists relaxed and fingers naturally extended. Between phrases she adds small directional changes: a quarter turn away from the lens and back, a step that closes distance with the camera, a shift of gaze that returns to the lens on the downbeat. The camera holds a Static Shot through the opening beats, then performs a slow Push In with small amplitude at slow speed, tightening from medium-full to medium while keeping her head and hands inside the frame at every point of the routine. Her expression stays confident and playful, with a light smile and eyes that meet the lens at the end of each phrase. The fabric of the transferred outfit swings and settles naturally with each step, turn, and arm extension, and her hair carries the momentum of her head movement half a beat behind it. The pacing, phrasing, and camera rhythm follow the structure established in <Video 1>: the routine builds energy through the middle of the clip, reaches its widest and fastest movement around the two-thirds point, then settles back toward stillness. The closing beats return her to a stance, framing, and lighting that closely match her opening position, so the clip loops cleanly for repeated live playback. No text, graphics, watermarks, or overlays appear anywhere in the frame at any point.\n\noverall_soundscape:\nSoft ambient room tone of an enclosed studio sits under the music, with faint fabric rustle as <Subject 1> moves and light rhythmic footfall on the floor as her weight transfers. No speech, no crowd, and no off-screen sound events occur at any point.\n\nnon_diegetic_music:\nAn upbeat instrumental track drives the routine from the first frame to the last; its tempo, instrumentation, and dynamics are copied verbatim from <Audio 1>. Its accents align with the weight transfers and arm extensions of <Subject 1>, and its energy build through the middle of the clip matches the build in her movement.",
-    "width": 480,
-    "height": 832,
-    "duration": 10,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788326871833,
-    "lastPlayedAt": 1788330350299,
-    "playCount": 109
-  },
-  {
-    "id": "40961b8b19f3",
-    "source": "h3console",
-    "edited": true,
-    "channel": "dance",
-    "characterId": "4a4dc93b-4984-495b-a660-535ce3d3fc51",
-    "garmentId": "6bbacecd-e492-4ec3-a9de-93744cea4b42",
-    "refDanceVideoId": "7b8e13ce-c9e3-4a19-bccc-5f54e6aae98e",
-    "backend": "h3_console",
-    "priority": "normal",
-    "requestedByUserId": null,
-    "prompt": "验证-人物 × 黑色旗袍",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the young woman whose visual identity is defined by <Picture 1>: 年轻女性，长直黑发，鹅蛋脸，气质清冷. Preserve her face, hairstyle, and body exactly as shown; only her outfit, pose, motion, and framing change in the target video.\n<Picture 1> is the character reference frame for <Subject 1>, showing her face, hairstyle, and body in a neutral standing pose, wearing a neutral outfit that does not appear in the target video; her footwear as shown here is retained unchanged in the target video.\n<Picture 2> is the garment reference image: 黑色旗袍，银色滚边，盘扣，侧开衩, laid flat with no person wearing it. Only its garment attributes (silhouette, color, fabric, trim, cut) transfer onto <Subject 1>; no other content of this image appears in the target video, and it does not affect her footwear.\n<Video 1> is the reference choreography clip. It defines the dance structure, phrasing, and camera rhythm that the target video follows; it is not a visual identity source, and none of its performer, background, or imagery appears in the target video.\n<Audio 1> is the background music track carried on <Video 1>, reused verbatim as the music bed of the target video. It contains no speech.\n\nsummary:\n[reference generation + attribute transfer + audio reuse] The target video shows <Subject 1>, now wearing the outfit transferred from <Picture 2>, performing 宽马步，双臂交替甩动，配合甩发动作，节奏感强的hip-hop风格 in a vertical studio setting, following the pacing and camera rhythm established in <Video 1>, with the music of <Audio 1> reused as its track.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - her face, hairstyle, and body are retained unchanged throughout; her outfit changes to the garment shown in <Picture 2>, her footwear stays as shown in <Picture 1>.\n<Picture 1>: fully_preserved - it supplies the face, hairstyle, body, and footwear of <Subject 1> and the opening framing of the shot.\n<Picture 2>: attribute_transfer - only the garment's silhouette, color, fabric, and trim transfer onto <Subject 1>; the image itself does not appear.\n<Video 1>: weak_reference - only the choreography phrasing and the camera rhythm are referenced; no performer, background, or imagery from it is retained.\n<Audio 1>: fully_copy - the music track is reused verbatim as the non-diegetic music bed.\n\ndetailed_description:\nA vibrant, mobile-first vertical dance video with warm, glossy studio lighting, saturated color, and a clean contemporary look sustained from the first frame to the last. [Shot 1] The shot begins from <Picture 1>, with <Subject 1> now wearing the outfit transferred from <Picture 2>. She stands centered in a medium-full vertical frame against a softly lit seamless backdrop that falls off into gentle shadow toward the edges. A warm key light sits slightly camera-left, a cooler rim light separates her shoulders and hair from the background, and a soft fill keeps her face evenly exposed. She begins in a relaxed neutral stance, weight settled on her back foot, hands loose at her sides, chin level with the lens. On the first musical accent she starts 宽马步，双臂交替甩动，配合甩发动作，节奏感强的hip-hop风格, and the routine unfolds continuously from there with no cut. Her weight transfers smoothly from foot to foot on the beat, hips leading each transfer and shoulders following a fraction behind, so the movement reads as one connected chain rather than a series of separate poses. Her arms trace clean, unhurried arcs that extend fully at the peak of each phrase and soften on the release, wrists relaxed and fingers naturally extended. Between phrases she adds small directional changes: a quarter turn away from the lens and back, a step that closes distance with the camera, a shift of gaze that returns to the lens on the downbeat. The camera holds a Static Shot through the opening beats, then performs a slow Push In with small amplitude at slow speed, tightening from medium-full to medium while keeping her head and hands inside the frame at every point of the routine. Her expression stays confident and playful, with a light smile and eyes that meet the lens at the end of each phrase. The fabric of the transferred outfit swings and settles naturally with each step, turn, and arm extension, and her hair carries the momentum of her head movement half a beat behind it. The pacing, phrasing, and camera rhythm follow the structure established in <Video 1>: the routine builds energy through the middle of the clip, reaches its widest and fastest movement around the two-thirds point, then settles back toward stillness. The closing beats return her to a stance, framing, and lighting that closely match her opening position, so the clip loops cleanly for repeated live playback. No text, graphics, watermarks, or overlays appear anywhere in the frame at any point.\n\noverall_soundscape:\nSoft ambient room tone of an enclosed studio sits under the music, with faint fabric rustle as <Subject 1> moves and light rhythmic footfall on the floor as her weight transfers. No speech, no crowd, and no off-screen sound events occur at any point.\n\nnon_diegetic_music:\nAn upbeat instrumental track drives the routine from the first frame to the last; its tempo, instrumentation, and dynamics are copied verbatim from <Audio 1>. Its accents align with the weight transfers and arm extensions of <Subject 1>, and its energy build through the middle of the clip matches the build in her movement.",
-    "width": 480,
-    "height": 832,
-    "duration": 10,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788325715238,
-    "lastPlayedAt": 1788330428372,
-    "playCount": 114
-  },
-  {
-    "id": "563b53b5bc79",
-    "source": "h3console",
-    "edited": true,
-    "channel": "dance",
-    "characterId": "4a4dc93b-4984-495b-a660-535ce3d3fc51",
-    "garmentId": "01793e15-70a0-4967-a5c7-b97eda799f95",
-    "refDanceVideoId": "e5e56a4f-2318-4fec-91be-fd94cf5dcdfc",
-    "backend": "h3_console",
-    "priority": "normal",
-    "requestedByUserId": null,
-    "prompt": "验证-人物 × 洋红色旗袍",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the young woman whose visual identity is defined by <Picture 1>: 年轻女性，长直黑发，鹅蛋脸，气质清冷. Preserve her face, hairstyle, and body exactly as shown; only her outfit, pose, motion, and framing change in the target video.\n<Picture 1> is the character reference frame for <Subject 1>, showing her face, hairstyle, and body in a neutral standing pose, wearing a neutral outfit that does not appear in the target video; her footwear as shown here is retained unchanged in the target video.\n<Picture 2> is the garment reference image: 洋红色旗袍，金色滚边，盘扣，侧开衩, laid flat with no person wearing it. Only its garment attributes (silhouette, color, fabric, trim, cut) transfer onto <Subject 1>; no other content of this image appears in the target video, and it does not affect her footwear.\n<Video 1> is the reference choreography clip. It defines the dance structure, phrasing, and camera rhythm that the target video follows; it is not a visual identity source, and none of its performer, background, or imagery appears in the target video.\n<Audio 1> is the background music track carried on <Video 1>, reused verbatim as the music bed of the target video. It contains no speech.\n\nsummary:\n[reference generation + attribute transfer + audio reuse] The target video shows <Subject 1>, now wearing the outfit transferred from <Picture 2>, performing 原地小幅度手势舞动，柔和克制的比心和摆手动作，节奏偏慢 in a vertical studio setting, following the pacing and camera rhythm established in <Video 1>, with the music of <Audio 1> reused as its track.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - her face, hairstyle, and body are retained unchanged throughout; her outfit changes to the garment shown in <Picture 2>, her footwear stays as shown in <Picture 1>.\n<Picture 1>: fully_preserved - it supplies the face, hairstyle, body, and footwear of <Subject 1> and the opening framing of the shot.\n<Picture 2>: attribute_transfer - only the garment's silhouette, color, fabric, and trim transfer onto <Subject 1>; the image itself does not appear.\n<Video 1>: weak_reference - only the choreography phrasing and the camera rhythm are referenced; no performer, background, or imagery from it is retained.\n<Audio 1>: fully_copy - the music track is reused verbatim as the non-diegetic music bed.\n\ndetailed_description:\nA vibrant, mobile-first vertical dance video with warm, glossy studio lighting, saturated color, and a clean contemporary look sustained from the first frame to the last. [Shot 1] The shot begins from <Picture 1>, with <Subject 1> now wearing the outfit transferred from <Picture 2>. She stands centered in a medium-full vertical frame against a softly lit seamless backdrop that falls off into gentle shadow toward the edges. A warm key light sits slightly camera-left, a cooler rim light separates her shoulders and hair from the background, and a soft fill keeps her face evenly exposed. She begins in a relaxed neutral stance, weight settled on her back foot, hands loose at her sides, chin level with the lens. On the first musical accent she starts 原地小幅度手势舞动，柔和克制的比心和摆手动作，节奏偏慢, and the routine unfolds continuously from there with no cut. Her weight transfers smoothly from foot to foot on the beat, hips leading each transfer and shoulders following a fraction behind, so the movement reads as one connected chain rather than a series of separate poses. Her arms trace clean, unhurried arcs that extend fully at the peak of each phrase and soften on the release, wrists relaxed and fingers naturally extended. Between phrases she adds small directional changes: a quarter turn away from the lens and back, a step that closes distance with the camera, a shift of gaze that returns to the lens on the downbeat. The camera holds a Static Shot through the opening beats, then performs a slow Push In with small amplitude at slow speed, tightening from medium-full to medium while keeping her head and hands inside the frame at every point of the routine. Her expression stays confident and playful, with a light smile and eyes that meet the lens at the end of each phrase. The fabric of the transferred outfit swings and settles naturally with each step, turn, and arm extension, and her hair carries the momentum of her head movement half a beat behind it. The pacing, phrasing, and camera rhythm follow the structure established in <Video 1>: the routine builds energy through the middle of the clip, reaches its widest and fastest movement around the two-thirds point, then settles back toward stillness. The closing beats return her to a stance, framing, and lighting that closely match her opening position, so the clip loops cleanly for repeated live playback. No text, graphics, watermarks, or overlays appear anywhere in the frame at any point.\n\noverall_soundscape:\nSoft ambient room tone of an enclosed studio sits under the music, with faint fabric rustle as <Subject 1> moves and light rhythmic footfall on the floor as her weight transfers. No speech, no crowd, and no off-screen sound events occur at any point.\n\nnon_diegetic_music:\nAn upbeat instrumental track drives the routine from the first frame to the last; its tempo, instrumentation, and dynamics are copied verbatim from <Audio 1>. Its accents align with the weight transfers and arm extensions of <Subject 1>, and its energy build through the middle of the clip matches the build in her movement.",
-    "width": 480,
-    "height": 832,
-    "duration": 10,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788324551855,
-    "lastPlayedAt": 1788330448390,
-    "playCount": 116
-  },
-  {
-    "id": "1014cfff97bd",
-    "source": "h3console",
-    "edited": true,
-    "channel": "dance",
-    "characterId": "4a4dc93b-4984-495b-a660-535ce3d3fc51",
-    "garmentId": "4bc86203-7854-47ea-b3fa-c38dbad5ad8b",
-    "refDanceVideoId": "e5e56a4f-2318-4fec-91be-fd94cf5dcdfc",
-    "backend": "h3_console",
-    "priority": "normal",
-    "requestedByUserId": null,
-    "prompt": "验证-人物 × 玫红色旗袍",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the young woman whose visual identity is defined by <Picture 1>: 年轻女性，长直黑发，鹅蛋脸，气质清冷. Preserve her face, hairstyle, and body exactly as shown; only her outfit, pose, motion, and framing change in the target video.\n<Picture 1> is the character reference frame for <Subject 1>, showing her face, hairstyle, and body in a neutral standing pose, wearing a neutral outfit that does not appear in the target video; her footwear as shown here is retained unchanged in the target video.\n<Picture 2> is the garment reference image: 玫红色旗袍，金色滚边，盘扣，侧开衩, laid flat with no person wearing it. Only its garment attributes (silhouette, color, fabric, trim, cut) transfer onto <Subject 1>; no other content of this image appears in the target video, and it does not affect her footwear.\n<Video 1> is the reference choreography clip. It defines the dance structure, phrasing, and camera rhythm that the target video follows; it is not a visual identity source, and none of its performer, background, or imagery appears in the target video.\n<Audio 1> is the background music track carried on <Video 1>, reused verbatim as the music bed of the target video. It contains no speech.\n\nsummary:\n[reference generation + attribute transfer + audio reuse] The target video shows <Subject 1>, now wearing the outfit transferred from <Picture 2>, performing 原地小幅度手势舞动，柔和克制的比心和摆手动作，节奏偏慢 in a vertical studio setting, following the pacing and camera rhythm established in <Video 1>, with the music of <Audio 1> reused as its track.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - her face, hairstyle, and body are retained unchanged throughout; her outfit changes to the garment shown in <Picture 2>, her footwear stays as shown in <Picture 1>.\n<Picture 1>: fully_preserved - it supplies the face, hairstyle, body, and footwear of <Subject 1> and the opening framing of the shot.\n<Picture 2>: attribute_transfer - only the garment's silhouette, color, fabric, and trim transfer onto <Subject 1>; the image itself does not appear.\n<Video 1>: weak_reference - only the choreography phrasing and the camera rhythm are referenced; no performer, background, or imagery from it is retained.\n<Audio 1>: fully_copy - the music track is reused verbatim as the non-diegetic music bed.\n\ndetailed_description:\nA vibrant, mobile-first vertical dance video with warm, glossy studio lighting, saturated color, and a clean contemporary look sustained from the first frame to the last. [Shot 1] The shot begins from <Picture 1>, with <Subject 1> now wearing the outfit transferred from <Picture 2>. She stands centered in a medium-full vertical frame against a softly lit seamless backdrop that falls off into gentle shadow toward the edges. A warm key light sits slightly camera-left, a cooler rim light separates her shoulders and hair from the background, and a soft fill keeps her face evenly exposed. She begins in a relaxed neutral stance, weight settled on her back foot, hands loose at her sides, chin level with the lens. On the first musical accent she starts 原地小幅度手势舞动，柔和克制的比心和摆手动作，节奏偏慢, and the routine unfolds continuously from there with no cut. Her weight transfers smoothly from foot to foot on the beat, hips leading each transfer and shoulders following a fraction behind, so the movement reads as one connected chain rather than a series of separate poses. Her arms trace clean, unhurried arcs that extend fully at the peak of each phrase and soften on the release, wrists relaxed and fingers naturally extended. Between phrases she adds small directional changes: a quarter turn away from the lens and back, a step that closes distance with the camera, a shift of gaze that returns to the lens on the downbeat. The camera holds a Static Shot through the opening beats, then performs a slow Push In with small amplitude at slow speed, tightening from medium-full to medium while keeping her head and hands inside the frame at every point of the routine. Her expression stays confident and playful, with a light smile and eyes that meet the lens at the end of each phrase. The fabric of the transferred outfit swings and settles naturally with each step, turn, and arm extension, and her hair carries the momentum of her head movement half a beat behind it. The pacing, phrasing, and camera rhythm follow the structure established in <Video 1>: the routine builds energy through the middle of the clip, reaches its widest and fastest movement around the two-thirds point, then settles back toward stillness. The closing beats return her to a stance, framing, and lighting that closely match her opening position, so the clip loops cleanly for repeated live playback. No text, graphics, watermarks, or overlays appear anywhere in the frame at any point.\n\noverall_soundscape:\nSoft ambient room tone of an enclosed studio sits under the music, with faint fabric rustle as <Subject 1> moves and light rhythmic footfall on the floor as her weight transfers. No speech, no crowd, and no off-screen sound events occur at any point.\n\nnon_diegetic_music:\nAn upbeat instrumental track drives the routine from the first frame to the last; its tempo, instrumentation, and dynamics are copied verbatim from <Audio 1>. Its accents align with the weight transfers and arm extensions of <Subject 1>, and its energy build through the middle of the clip matches the build in her movement.",
-    "width": 480,
-    "height": 832,
-    "duration": 10,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788323385790,
-    "lastPlayedAt": 1788330468401,
-    "playCount": 118
-  },
-  {
-    "id": "fb855e11be4d",
-    "source": "h3console",
-    "edited": true,
-    "channel": "dance",
-    "characterId": "4a4dc93b-4984-495b-a660-535ce3d3fc51",
-    "garmentId": "a49b0a3d-3cff-4730-a941-d1621666e203",
-    "refDanceVideoId": "7b8e13ce-c9e3-4a19-bccc-5f54e6aae98e",
-    "backend": "h3_console",
-    "priority": "normal",
-    "requestedByUserId": null,
-    "prompt": "验证-人物 × 青色旗袍",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the young woman whose visual identity is defined by <Picture 1>: 年轻女性，长直黑发，鹅蛋脸，气质清冷. Preserve her face, hairstyle, and body exactly as shown; only her outfit, pose, motion, and framing change in the target video.\n<Picture 1> is the character reference frame for <Subject 1>, showing her face, hairstyle, and body in a neutral standing pose, wearing a neutral outfit that does not appear in the target video; her footwear as shown here is retained unchanged in the target video.\n<Picture 2> is the garment reference image: 青色旗袍，浅紫滚边，盘扣，侧开衩, laid flat with no person wearing it. Only its garment attributes (silhouette, color, fabric, trim, cut) transfer onto <Subject 1>; no other content of this image appears in the target video, and it does not affect her footwear.\n<Video 1> is the reference choreography clip. It defines the dance structure, phrasing, and camera rhythm that the target video follows; it is not a visual identity source, and none of its performer, background, or imagery appears in the target video.\n<Audio 1> is the background music track carried on <Video 1>, reused verbatim as the music bed of the target video. It contains no speech.\n\nsummary:\n[reference generation + attribute transfer + audio reuse] The target video shows <Subject 1>, now wearing the outfit transferred from <Picture 2>, performing 宽马步，双臂交替甩动，配合甩发动作，节奏感强的hip-hop风格 in a vertical studio setting, following the pacing and camera rhythm established in <Video 1>, with the music of <Audio 1> reused as its track.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - her face, hairstyle, and body are retained unchanged throughout; her outfit changes to the garment shown in <Picture 2>, her footwear stays as shown in <Picture 1>.\n<Picture 1>: fully_preserved - it supplies the face, hairstyle, body, and footwear of <Subject 1> and the opening framing of the shot.\n<Picture 2>: attribute_transfer - only the garment's silhouette, color, fabric, and trim transfer onto <Subject 1>; the image itself does not appear.\n<Video 1>: weak_reference - only the choreography phrasing and the camera rhythm are referenced; no performer, background, or imagery from it is retained.\n<Audio 1>: fully_copy - the music track is reused verbatim as the non-diegetic music bed.\n\ndetailed_description:\nA vibrant, mobile-first vertical dance video with warm, glossy studio lighting, saturated color, and a clean contemporary look sustained from the first frame to the last. [Shot 1] The shot begins from <Picture 1>, with <Subject 1> now wearing the outfit transferred from <Picture 2>. She stands centered in a medium-full vertical frame against a softly lit seamless backdrop that falls off into gentle shadow toward the edges. A warm key light sits slightly camera-left, a cooler rim light separates her shoulders and hair from the background, and a soft fill keeps her face evenly exposed. She begins in a relaxed neutral stance, weight settled on her back foot, hands loose at her sides, chin level with the lens. On the first musical accent she starts 宽马步，双臂交替甩动，配合甩发动作，节奏感强的hip-hop风格, and the routine unfolds continuously from there with no cut. Her weight transfers smoothly from foot to foot on the beat, hips leading each transfer and shoulders following a fraction behind, so the movement reads as one connected chain rather than a series of separate poses. Her arms trace clean, unhurried arcs that extend fully at the peak of each phrase and soften on the release, wrists relaxed and fingers naturally extended. Between phrases she adds small directional changes: a quarter turn away from the lens and back, a step that closes distance with the camera, a shift of gaze that returns to the lens on the downbeat. The camera holds a Static Shot through the opening beats, then performs a slow Push In with small amplitude at slow speed, tightening from medium-full to medium while keeping her head and hands inside the frame at every point of the routine. Her expression stays confident and playful, with a light smile and eyes that meet the lens at the end of each phrase. The fabric of the transferred outfit swings and settles naturally with each step, turn, and arm extension, and her hair carries the momentum of her head movement half a beat behind it. The pacing, phrasing, and camera rhythm follow the structure established in <Video 1>: the routine builds energy through the middle of the clip, reaches its widest and fastest movement around the two-thirds point, then settles back toward stillness. The closing beats return her to a stance, framing, and lighting that closely match her opening position, so the clip loops cleanly for repeated live playback. No text, graphics, watermarks, or overlays appear anywhere in the frame at any point.\n\noverall_soundscape:\nSoft ambient room tone of an enclosed studio sits under the music, with faint fabric rustle as <Subject 1> moves and light rhythmic footfall on the floor as her weight transfers. No speech, no crowd, and no off-screen sound events occur at any point.\n\nnon_diegetic_music:\nAn upbeat instrumental track drives the routine from the first frame to the last; its tempo, instrumentation, and dynamics are copied verbatim from <Audio 1>. Its accents align with the weight transfers and arm extensions of <Subject 1>, and its energy build through the middle of the clip matches the build in her movement.",
-    "width": 480,
-    "height": 832,
-    "duration": 10,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788322206869,
-    "lastPlayedAt": 1788330528445,
-    "playCount": 118
-  },
-  {
-    "id": "60bf07534b3a",
-    "source": "h3console",
-    "edited": true,
-    "channel": "dance",
-    "characterId": "4a4dc93b-4984-495b-a660-535ce3d3fc51",
-    "garmentId": "2c4487a8-6ff7-402c-8e13-dfefcd8eda6f",
-    "refDanceVideoId": "7b8e13ce-c9e3-4a19-bccc-5f54e6aae98e",
-    "backend": "h3_console",
-    "priority": "normal",
-    "requestedByUserId": null,
-    "prompt": "验证-人物 × 橙色旗袍",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the young woman whose visual identity is defined by <Picture 1>: 年轻女性，长直黑发，鹅蛋脸，气质清冷. Preserve her face, hairstyle, and body exactly as shown; only her outfit, pose, motion, and framing change in the target video.\n<Picture 1> is the character reference frame for <Subject 1>, showing her face, hairstyle, and body in a neutral standing pose, wearing a neutral outfit that does not appear in the target video; her footwear as shown here is retained unchanged in the target video.\n<Picture 2> is the garment reference image: 橙色旗袍，紫色滚边，盘扣，侧开衩, laid flat with no person wearing it. Only its garment attributes (silhouette, color, fabric, trim, cut) transfer onto <Subject 1>; no other content of this image appears in the target video, and it does not affect her footwear.\n<Video 1> is the reference choreography clip. It defines the dance structure, phrasing, and camera rhythm that the target video follows; it is not a visual identity source, and none of its performer, background, or imagery appears in the target video.\n<Audio 1> is the background music track carried on <Video 1>, reused verbatim as the music bed of the target video. It contains no speech.\n\nsummary:\n[reference generation + attribute transfer + audio reuse] The target video shows <Subject 1>, now wearing the outfit transferred from <Picture 2>, performing 宽马步，双臂交替甩动，配合甩发动作，节奏感强的hip-hop风格 in a vertical studio setting, following the pacing and camera rhythm established in <Video 1>, with the music of <Audio 1> reused as its track.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - her face, hairstyle, and body are retained unchanged throughout; her outfit changes to the garment shown in <Picture 2>, her footwear stays as shown in <Picture 1>.\n<Picture 1>: fully_preserved - it supplies the face, hairstyle, body, and footwear of <Subject 1> and the opening framing of the shot.\n<Picture 2>: attribute_transfer - only the garment's silhouette, color, fabric, and trim transfer onto <Subject 1>; the image itself does not appear.\n<Video 1>: weak_reference - only the choreography phrasing and the camera rhythm are referenced; no performer, background, or imagery from it is retained.\n<Audio 1>: fully_copy - the music track is reused verbatim as the non-diegetic music bed.\n\ndetailed_description:\nA vibrant, mobile-first vertical dance video with warm, glossy studio lighting, saturated color, and a clean contemporary look sustained from the first frame to the last. [Shot 1] The shot begins from <Picture 1>, with <Subject 1> now wearing the outfit transferred from <Picture 2>. She stands centered in a medium-full vertical frame against a softly lit seamless backdrop that falls off into gentle shadow toward the edges. A warm key light sits slightly camera-left, a cooler rim light separates her shoulders and hair from the background, and a soft fill keeps her face evenly exposed. She begins in a relaxed neutral stance, weight settled on her back foot, hands loose at her sides, chin level with the lens. On the first musical accent she starts 宽马步，双臂交替甩动，配合甩发动作，节奏感强的hip-hop风格, and the routine unfolds continuously from there with no cut. Her weight transfers smoothly from foot to foot on the beat, hips leading each transfer and shoulders following a fraction behind, so the movement reads as one connected chain rather than a series of separate poses. Her arms trace clean, unhurried arcs that extend fully at the peak of each phrase and soften on the release, wrists relaxed and fingers naturally extended. Between phrases she adds small directional changes: a quarter turn away from the lens and back, a step that closes distance with the camera, a shift of gaze that returns to the lens on the downbeat. The camera holds a Static Shot through the opening beats, then performs a slow Push In with small amplitude at slow speed, tightening from medium-full to medium while keeping her head and hands inside the frame at every point of the routine. Her expression stays confident and playful, with a light smile and eyes that meet the lens at the end of each phrase. The fabric of the transferred outfit swings and settles naturally with each step, turn, and arm extension, and her hair carries the momentum of her head movement half a beat behind it. The pacing, phrasing, and camera rhythm follow the structure established in <Video 1>: the routine builds energy through the middle of the clip, reaches its widest and fastest movement around the two-thirds point, then settles back toward stillness. The closing beats return her to a stance, framing, and lighting that closely match her opening position, so the clip loops cleanly for repeated live playback. No text, graphics, watermarks, or overlays appear anywhere in the frame at any point.\n\noverall_soundscape:\nSoft ambient room tone of an enclosed studio sits under the music, with faint fabric rustle as <Subject 1> moves and light rhythmic footfall on the floor as her weight transfers. No speech, no crowd, and no off-screen sound events occur at any point.\n\nnon_diegetic_music:\nAn upbeat instrumental track drives the routine from the first frame to the last; its tempo, instrumentation, and dynamics are copied verbatim from <Audio 1>. Its accents align with the weight transfers and arm extensions of <Subject 1>, and its energy build through the middle of the clip matches the build in her movement.",
-    "width": 480,
-    "height": 832,
-    "duration": 10,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788321031897,
-    "lastPlayedAt": 1788330488413,
-    "playCount": 118
-  },
-  {
-    "id": "c529d308370a",
-    "source": "h3console",
-    "edited": true,
-    "channel": "dance",
-    "characterId": "4a4dc93b-4984-495b-a660-535ce3d3fc51",
-    "garmentId": "ac39a43f-6e81-4d2a-8fdb-af9c4d686e44",
-    "refDanceVideoId": "f2c6304c-1155-4d82-afe9-17de85c9398a",
-    "backend": "h3_console",
-    "priority": "normal",
-    "requestedByUserId": null,
-    "prompt": "验证-人物 × 验证-旗袍",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the young woman whose visual identity is defined by <Picture 1>: 年轻女性，长直黑发，鹅蛋脸，气质清冷. Preserve her face, hairstyle, and body exactly as shown; only her outfit, pose, motion, and framing change in the target video.\n<Picture 1> is the character reference frame for <Subject 1>, showing her face, hairstyle, and body in a neutral standing pose, wearing a neutral outfit that does not appear in the target video; her footwear as shown here is retained unchanged in the target video.\n<Picture 2> is the garment reference image: 红色旗袍，金色滚边，侧开衩, laid flat with no person wearing it. Only its garment attributes (silhouette, color, fabric, trim, cut) transfer onto <Subject 1>; no other content of this image appears in the target video, and it does not affect her footwear.\n<Video 1> is the reference choreography clip. It defines the dance structure, phrasing, and camera rhythm that the target video follows; it is not a visual identity source, and none of its performer, background, or imagery appears in the target video.\n<Audio 1> is the background music track carried on <Video 1>, reused verbatim as the music bed of the target video. It contains no speech.\n\nsummary:\n[reference generation + attribute transfer + audio reuse] The target video shows <Subject 1>, now wearing the outfit transferred from <Picture 2>, performing 宽马步，甩臂过顶，比心，hip-hop风格能量感 in a vertical studio setting, following the pacing and camera rhythm established in <Video 1>, with the music of <Audio 1> reused as its track.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - her face, hairstyle, and body are retained unchanged throughout; her outfit changes to the garment shown in <Picture 2>, her footwear stays as shown in <Picture 1>.\n<Picture 1>: fully_preserved - it supplies the face, hairstyle, body, and footwear of <Subject 1> and the opening framing of the shot.\n<Picture 2>: attribute_transfer - only the garment's silhouette, color, fabric, and trim transfer onto <Subject 1>; the image itself does not appear.\n<Video 1>: weak_reference - only the choreography phrasing and the camera rhythm are referenced; no performer, background, or imagery from it is retained.\n<Audio 1>: fully_copy - the music track is reused verbatim as the non-diegetic music bed.\n\ndetailed_description:\nA vibrant, mobile-first vertical dance video with warm, glossy studio lighting, saturated color, and a clean contemporary look sustained from the first frame to the last. [Shot 1] The shot begins from <Picture 1>, with <Subject 1> now wearing the outfit transferred from <Picture 2>. She stands centered in a medium-full vertical frame against a softly lit seamless backdrop that falls off into gentle shadow toward the edges. A warm key light sits slightly camera-left, a cooler rim light separates her shoulders and hair from the background, and a soft fill keeps her face evenly exposed. She begins in a relaxed neutral stance, weight settled on her back foot, hands loose at her sides, chin level with the lens. On the first musical accent she starts 宽马步，甩臂过顶，比心，hip-hop风格能量感, and the routine unfolds continuously from there with no cut. Her weight transfers smoothly from foot to foot on the beat, hips leading each transfer and shoulders following a fraction behind, so the movement reads as one connected chain rather than a series of separate poses. Her arms trace clean, unhurried arcs that extend fully at the peak of each phrase and soften on the release, wrists relaxed and fingers naturally extended. Between phrases she adds small directional changes: a quarter turn away from the lens and back, a step that closes distance with the camera, a shift of gaze that returns to the lens on the downbeat. The camera holds a Static Shot through the opening beats, then performs a slow Push In with small amplitude at slow speed, tightening from medium-full to medium while keeping her head and hands inside the frame at every point of the routine. Her expression stays confident and playful, with a light smile and eyes that meet the lens at the end of each phrase. The fabric of the transferred outfit swings and settles naturally with each step, turn, and arm extension, and her hair carries the momentum of her head movement half a beat behind it. The pacing, phrasing, and camera rhythm follow the structure established in <Video 1>: the routine builds energy through the middle of the clip, reaches its widest and fastest movement around the two-thirds point, then settles back toward stillness. The closing beats return her to a stance, framing, and lighting that closely match her opening position, so the clip loops cleanly for repeated live playback. No text, graphics, watermarks, or overlays appear anywhere in the frame at any point.\n\noverall_soundscape:\nSoft ambient room tone of an enclosed studio sits under the music, with faint fabric rustle as <Subject 1> moves and light rhythmic footfall on the floor as her weight transfers. No speech, no crowd, and no off-screen sound events occur at any point.\n\nnon_diegetic_music:\nAn upbeat instrumental track drives the routine from the first frame to the last; its tempo, instrumentation, and dynamics are copied verbatim from <Audio 1>. Its accents align with the weight transfers and arm extensions of <Subject 1>, and its energy build through the middle of the clip matches the build in her movement.",
-    "width": 480,
-    "height": 832,
-    "duration": 10,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788319853705,
-    "lastPlayedAt": 1788330508424,
-    "playCount": 118
-  },
-  {
-    "id": "f23903118436",
-    "source": "h3console",
-    "prompt": "验证-人物 × 验证-旗袍",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the young woman whose visual identity is defined by <Picture 1>: 年轻女性，长直黑发，鹅蛋脸，气质清冷. Preserve her face, hairstyle, and body exactly as shown; only her outfit, pose, motion, and framing change in the target video.\n<Picture 1> is the character reference frame for <Subject 1>, showing her face, hairstyle, and body in a neutral standing pose, wearing a neutral outfit that does not appear in the target video; her footwear as shown here is retained unchanged in the target video.\n<Picture 2> is the garment reference image: 红色旗袍，金色滚边，侧开衩, laid flat with no person wearing it. Only its garment attributes (silhouette, color, fabric, trim, cut) transfer onto <Subject 1>; no other content of this image appears in the target video, and it does not affect her footwear.\n<Video 1> is the reference choreography clip. It defines the dance structure, phrasing, and camera rhythm that the target video follows; it is not a visual identity source, and none of its performer, background, or imagery appears in the target video.\n<Audio 1> is the background music track carried on <Video 1>, reused verbatim as the music bed of the target video. It contains no speech.\n\nsummary:\n[reference generation + attribute transfer + audio reuse] The target video shows <Subject 1>, now wearing the outfit transferred from <Picture 2>, performing 宽马步，甩臂过顶，比心，hip-hop风格能量感 in a vertical studio setting, following the pacing and camera rhythm established in <Video 1>, with the music of <Audio 1> reused as its track.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - her face, hairstyle, and body are retained unchanged throughout; her outfit changes to the garment shown in <Picture 2>, her footwear stays as shown in <Picture 1>.\n<Picture 1>: fully_preserved - it supplies the face, hairstyle, body, and footwear of <Subject 1> and the opening framing of the shot.\n<Picture 2>: attribute_transfer - only the garment's silhouette, color, fabric, and trim transfer onto <Subject 1>; the image itself does not appear.\n<Video 1>: weak_reference - only the choreography phrasing and the camera rhythm are referenced; no performer, background, or imagery from it is retained.\n<Audio 1>: fully_copy - the music track is reused verbatim as the non-diegetic music bed.\n\ndetailed_description:\nA vibrant, mobile-first vertical dance video with warm, glossy studio lighting, saturated color, and a clean contemporary look sustained from the first frame to the last. [Shot 1] The shot begins from <Picture 1>, with <Subject 1> now wearing the outfit transferred from <Picture 2>. She stands centered in a medium-full vertical frame against a softly lit seamless backdrop that falls off into gentle shadow toward the edges. A warm key light sits slightly camera-left, a cooler rim light separates her shoulders and hair from the background, and a soft fill keeps her face evenly exposed. She begins in a relaxed neutral stance, weight settled on her back foot, hands loose at her sides, chin level with the lens. On the first musical accent she starts 宽马步，甩臂过顶，比心，hip-hop风格能量感, and the routine unfolds continuously from there with no cut. Her weight transfers smoothly from foot to foot on the beat, hips leading each transfer and shoulders following a fraction behind, so the movement reads as one connected chain rather than a series of separate poses. Her arms trace clean, unhurried arcs that extend fully at the peak of each phrase and soften on the release, wrists relaxed and fingers naturally extended. Between phrases she adds small directional changes: a quarter turn away from the lens and back, a step that closes distance with the camera, a shift of gaze that returns to the lens on the downbeat. The camera holds a Static Shot through the opening beats, then performs a slow Push In with small amplitude at slow speed, tightening from medium-full to medium while keeping her head and hands inside the frame at every point of the routine. Her expression stays confident and playful, with a light smile and eyes that meet the lens at the end of each phrase. The fabric of the transferred outfit swings and settles naturally with each step, turn, and arm extension, and her hair carries the momentum of her head movement half a beat behind it. The pacing, phrasing, and camera rhythm follow the structure established in <Video 1>: the routine builds energy through the middle of the clip, reaches its widest and fastest movement around the two-thirds point, then settles back toward stillness. The closing beats return her to a stance, framing, and lighting that closely match her opening position, so the clip loops cleanly for repeated live playback. No text, graphics, watermarks, or overlays appear anywhere in the frame at any point.\n\noverall_soundscape:\nSoft ambient room tone of an enclosed studio sits under the music, with faint fabric rustle as <Subject 1> moves and light rhythmic footfall on the floor as her weight transfers. No speech, no crowd, and no off-screen sound events occur at any point.\n\nnon_diegetic_music:\nAn upbeat instrumental track drives the routine from the first frame to the last; its tempo, instrumentation, and dynamics are copied verbatim from <Audio 1>. Its accents align with the weight transfers and arm extensions of <Subject 1>, and its energy build through the middle of the clip matches the build in her movement.",
-    "edited": true,
-    "width": 480,
-    "height": 832,
-    "duration": 10,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788317315021,
-    "channel": "dance",
-    "characterId": "4a4dc93b-4984-495b-a660-535ce3d3fc51",
-    "garmentId": "ac39a43f-6e81-4d2a-8fdb-af9c4d686e44",
-    "refDanceVideoId": "f2c6304c-1155-4d82-afe9-17de85c9398a",
-    "backend": "h3_console",
-    "priority": "normal",
-    "requestedByUserId": null,
-    "lastPlayedAt": 1788330568476,
-    "playCount": 118
-  },
-  {
-    "id": "b4b9b6568ea8",
-    "source": "h3console",
-    "prompt": "星际迷航穿越沙漠 主角是星际迷航的主人翁，背景是星际迷航某一个星球，主角的台词是大喊“为了荣耀，出发～～”，然后切换到千军万马的航空舰队，开始点火启动，",
-    "fullPrompt": "星际迷航穿越沙漠\n主角是星际迷航的主人翁，背景是星际迷航某一个星球，主角的台词是大喊“为了荣耀，出发～～”，然后切换到千军万马的航空舰队，开始点火启动，",
-    "edited": false,
-    "width": 480,
-    "height": 832,
-    "duration": 10.1,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1788264244312,
-    "lastPlayedAt": 1788331519241,
-    "playCount": 1
-  },
-  {
-    "id": "734b59af7b01",
-    "source": "h3console",
-    "prompt": "For the target video, at 0.00 seconds into the target video, (from [Shot 1]) is fully referenced. integrated_multimodal_",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nintegrated_multimodal_description:\n[Shot 1] The shot begins from <Picture 1>: a live-action mid-2000s office mockumentary frame with the same male regional manager in a dark suit, white shirt, and patterned tie standing beside the large Dunder Mifflin wall sign. Preserve the fluorescent office lighting, the handheld documentary camera feel, the venetian blinds, the beige wall, and the existing wall sign text. No subtitles, no captions, no UI overlays, no watermark, and no newly generated on-screen text. The camera holds with a slight natural wobble as the manager continues his confident office-tour pose. A small screw creak comes from the sign behind him; the top edge of the sign loosens and dips.\n[Shot 2] At 00:02.200, the sign suddenly drops partway down the wall with a plastic clatter. The handheld camera jolts slightly but stays in the same office corridor. The manager flinches, looks back, then instantly faces the camera and tries to look composed. The manager (S1) says: <d>[English] That was a test.</d> His expression insists that the accident was intentional.\n[Shot 3] At 00:04.100, the camera cuts to the receptionist at her desk, still in the same office. She looks over with practical concern, holding a pen near the phone. The receptionist (S2) says: <d>[English] Should I call maintenance?</d> The manager answers from off-screen while his lips are not visible: <d>[English] No, then it becomes real.</d>\n[Shot 4] At 00:06.000, the shot cuts to the salesman at his desk for a dry mockumentary reaction. He glances at the fallen sign, then gives the camera a tiny look. The salesman (S3) says: <d>[English] The sign quit first.</d> The timing is quiet and awkward, not broad parody.\n[Shot 5] At 00:07.500, the intense coworker steps in and presses the sign back against the wall too seriously. The manager grabs the other side but mostly gets in the way. The intense coworker (S4) says: <d>[English] Structural failure.</d> The manager (S1) says: <d>[English] Leadership opportunity.</d> Keep the action small and realistic: no injury, no shattered sign, no disaster.\n[Shot 6] At 00:09.600, the sign is nearly back in its original position. The receptionist quietly straightens one corner while the intense coworker holds pressure. The manager smooths his jacket, steps back beside the sign, and resets his proud tour posture. The manager (S1) says: <d>[English] Again. Same me, better gravity.</d>\n[Shot 7] At 00:11.300, use a tiny documentary-style jump cut back to a near match of <Picture 1>. The Dunder Mifflin sign is upright again, the manager stands beside it, and the camera settles into the original sign-visible composition by 00:12.250 so the source episode can resume from the same moment.\n\noverall_soundscape:\nNatural office room tone runs throughout: fluorescent hum, distant desk movement, a brief plastic sign clatter, quick tape or wall contact, and restrained workplace reactions. Dialogue is spoken in English only and must never appear as subtitles or captions.\n\nnon_diegetic_music:\nN/A",
-    "edited": false,
-    "width": 832,
-    "height": 480,
-    "duration": 12.3,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788258526737
-  },
-  {
-    "id": "e9e9907999a3",
-    "source": "h3console",
-    "prompt": "明日香大战老太太",
-    "fullPrompt": "明日香大战老太太",
-    "edited": false,
-    "width": 480,
-    "height": 832,
-    "duration": 5,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788254902903,
-    "lastPlayedAt": 1788331513250,
-    "playCount": 1
-  },
-  {
-    "id": "13baa0ba32c5",
-    "source": "h3console",
-    "prompt": "明日香大战老太太",
-    "fullPrompt": "明日香大战老太太",
-    "edited": false,
-    "width": 480,
-    "height": 832,
-    "duration": 5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1788254902838,
-    "lastPlayedAt": 1788331529248,
-    "playCount": 1
-  },
-  {
-    "id": "8d782b62dd8e",
-    "source": "h3console",
-    "prompt": "星际迷航穿越沙漠",
-    "fullPrompt": "星际迷航穿越沙漠",
-    "edited": false,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1788254437728
-  },
-  {
-    "id": "1d116e9a5424",
-    "source": "h3console",
-    "prompt": "为什么跑这么快",
-    "fullPrompt": "为什么跑这么快",
-    "edited": false,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1788254400661
-  },
-  {
-    "id": "4af4c61b64a6",
-    "source": "h3console",
-    "prompt": "一只白猫坐着，镜头缓慢推进，最终定格",
-    "fullPrompt": "一只白猫坐着，镜头缓慢推进，最终定格",
-    "edited": false,
-    "width": 832,
-    "height": 480,
-    "duration": 1,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788251288958,
-    "lastPlayedAt": 1788331513235,
-    "playCount": 1
-  },
-  {
-    "id": "8024972d6019",
-    "source": "h3console",
-    "prompt": "蛤蟆文太跳出来了打架",
-    "fullPrompt": "蛤蟆文太跳出来了打架",
-    "edited": false,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1788249485816,
-    "lastPlayedAt": 1788331503223,
-    "playCount": 1
-  },
-  {
-    "id": "0ade3648412f",
-    "source": "h3console",
-    "prompt": "来一只猫在草地上玩耍",
-    "fullPrompt": "来一只猫在草地上玩耍",
-    "edited": false,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1788239781137
-  },
-  {
-    "id": "af0d86637410",
-    "source": "h3console",
-    "prompt": "叫几个小猫咪和绿巨人一起跳舞",
-    "fullPrompt": "叫几个小猫咪和绿巨人一起跳舞",
-    "edited": false,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1788239123020,
-    "lastPlayedAt": 1788331507228,
-    "playCount": 1
-  },
-  {
-    "id": "aa1e9123bd35",
-    "source": "h3console",
-    "prompt": "道士打量古宅，一脸怀疑",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single twenty-one-year-old half-trained Chinese Taoist whose facial identity, tied hair, worn Taoist robe, cloth bundle, old talismans, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"suspicious_examining\". The state must be unmistakable from a single silent frame: superstitious scrutiny: focused eyes, a cautious frown, measured breathing, and the uncertainty of someone matching evidence against incomplete ritual knowledge. Its defining motion is: His eyes trace one detail outside the frame, his lips pause as if recalling a half-remembered rule, then he looks back with guarded suspicion. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: He occupies the right third from the upper chest upward, torso angled toward off-screen evidence, neck extended slightly forward, and chin lowered in concentration. He remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish superstitious scrutiny: focused eyes, a cautious frown, measured breathing, and the uncertainty of someone matching evidence against incomplete ritual knowledge. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into scholarly certainty, panic, attraction, confidence, disgust, exaggerated spell casting.\n\nHe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. His mouth moves quietly and deliberately, stopping completely between short phrases while he searches his memory. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. He produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. His eyes trace one detail outside the frame, his lips pause as if recalling a half-remembered rule, then he looks back with guarded suspicion. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788192916199,
-    "promptEn": "taoist eyes the old house suspiciously",
-    "lastPlayedAt": 1788330810648,
-    "playCount": 22
-  },
-  {
-    "id": "cb5baad5d724",
-    "source": "h3console",
-    "prompt": "书生警惕地观察房间",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single frail twenty-four-year-old Chinese scholar whose facial identity, tied dark hair, worn travel robe, old satchel, bookish appearance, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"wary_observing\". The state must be unmistakable from a single silent frame: quiet suspicion and analytical attention: narrowed eyes, a faintly furrowed brow, controlled breathing, and careful listening before each response. Its defining motion is: His gaze shifts once toward a suspicious object outside the frame, pauses as if comparing details, then returns to the off-screen listener without moving his feet. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: He occupies the right third of the frame from the upper chest upward, torso held in a three-quarter angle toward the suspicious object, with his head slightly forward. He remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish quiet suspicion and analytical attention: narrowed eyes, a faintly furrowed brow, controlled breathing, and careful listening before each response. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into panic, trust, infatuation, hostility, smugness, exaggerated detective gestures.\n\nHe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. His articulation is deliberate and precise, separated by long closed-mouth pauses used to inspect and reconsider details. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. He produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. His gaze shifts once toward a suspicious object outside the frame, pauses as if comparing details, then returns to the off-screen listener without moving his feet. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788192915845,
-    "promptEn": "scholar watches the room warily",
-    "lastPlayedAt": 1788330248202,
-    "playCount": 22
-  },
-  {
-    "id": "e8cbb73560ef",
-    "source": "h3console",
-    "prompt": "道士坚定地准备驱邪",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single twenty-one-year-old half-trained Chinese Taoist whose facial identity, tied hair, worn Taoist robe, cloth bundle, old talismans, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"resolute_exorcising\". The state must be unmistakable from a single silent frame: earned resolve despite lingering fear: steady eyes, a set jaw, controlled breathing, and concentration replacing earlier bluff. Its defining motion is: He gathers himself with one deliberate breath, fixes his gaze on the off-screen threat, and gives a small decisive nod before returning near the opening posture. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: He stands centered from mid-chest upward, torso square and upright, shoulders level, head vertical, and gaze aligned directly toward the threat. He remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish earned resolve despite lingering fear: steady eyes, a set jaw, controlled breathing, and concentration replacing earlier bluff. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into triumph, arrogance, rage, panic, supernatural hand effects, full-body ritual action.\n\nHe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. His lips move in a steady deliberate cadence with clear closures and measured pauses that preserve the set jaw. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. He produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. He gathers himself with one deliberate breath, fixes his gaze on the off-screen threat, and gives a small decisive nod before returning near the opening posture. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788192915549,
-    "promptEn": "taoist steels himself to exorcise something",
-    "lastPlayedAt": 1788330740592,
-    "playCount": 22
-  },
-  {
-    "id": "5b6dcb357792",
-    "source": "h3console",
-    "prompt": "书生态度软化，开始信任对方",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single frail twenty-four-year-old Chinese scholar whose facial identity, tied dark hair, worn travel robe, old satchel, bookish appearance, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"softened_trusting\". The state must be unmistakable from a single silent frame: cautious relief and emerging trust: the brow loosens, the shoulders settle, and the eyes become warmer without losing all vigilance. Its defining motion is: He releases a small held breath, lowers his chin in acknowledgement, and keeps steady eye contact with the off-screen listener before returning to his neutral posture. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: He stands centered from mid-chest upward, torso open toward the listener, shoulders lowered evenly, and head upright at a comfortable conversational distance. He remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish cautious relief and emerging trust: the brow loosens, the shoulders settle, and the eyes become warmer without losing all vigilance. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into romantic attraction, drowsiness, fear, suspicion, broad smiling, complete relaxation.\n\nHe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. His lips move in an even gentle rhythm with smooth transitions and relaxed pauses, without losing the cautious warmth. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. He produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. He releases a small held breath, lowers his chin in acknowledgement, and keeps steady eye contact with the off-screen listener before returning to his neutral posture. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788192915268,
-    "promptEn": "scholar softens up, starts to trust",
-    "lastPlayedAt": 1788328618938,
-    "playCount": 21
-  },
-  {
-    "id": "0c7565a5c8e8",
-    "source": "h3console",
-    "prompt": "道士嘴硬，其实很紧张",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single twenty-one-year-old half-trained Chinese Taoist whose facial identity, tied hair, worn Taoist robe, cloth bundle, old talismans, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"nervous_bravado\". The state must be unmistakable from a single silent frame: performative confidence over visible nerves: a raised chin, overly deliberate articulation, restless eyes, and tension that leaks through the pose. Its defining motion is: He straightens his shoulders to look authoritative, flicks one uncertain glance aside, then resumes eye contact as if hoping the bluff was convincing. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: He occupies the left third from mid-chest upward, chest pushed forward too deliberately, chin raised, and torso angled toward the listener as if performing authority. He remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish performative confidence over visible nerves: a raised chin, overly deliberate articulation, restless eyes, and tension that leaks through the pose. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into true mastery, calm certainty, panic, charm, rage, comic slapstick.\n\nHe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. He over-enunciates broad but controlled mouth shapes, then closes his lips abruptly whenever his confidence falters. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. He produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. He straightens his shoulders to look authoritative, flicks one uncertain glance aside, then resumes eye contact as if hoping the bluff was convincing. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788192914949,
-    "promptEn": "taoist acts tough but is nervous",
-    "lastPlayedAt": 1788326972039,
-    "playCount": 21
-  },
-  {
-    "id": "167ce1bfa285",
-    "source": "h3console",
-    "prompt": "书生害怕地想要逃跑",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single frail twenty-four-year-old Chinese scholar whose facial identity, tied dark hair, worn travel robe, old satchel, bookish appearance, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"frightened_fleeing\". The state must be unmistakable from a single silent frame: restrained exhaustion and active fear: slightly widened eyes, shallow breathing, damp tension around the jaw, and an urgent need to be believed. Its defining motion is: His eyes briefly check the doorway, his head pulls back a few centimeters, and one tense swallow passes before he returns his attention to the off-screen listener. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: He occupies the left third of the frame from mid-chest upward, torso turned twenty-five degrees away from the listener, shoulders drawn back, and head angled toward the doorway. He remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish restrained exhaustion and active fear: slightly widened eyes, shallow breathing, damp tension around the jaw, and an urgent need to be believed. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into relief, attraction, confidence, anger, smiling, theatrical panic.\n\nHe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. His lips form short uneven bursts with frequent closed-mouth pauses, while the jaw remains visibly tense. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. He produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. His eyes briefly check the doorway, his head pulls back a few centimeters, and one tense swallow passes before he returns his attention to the off-screen listener. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788192914047,
-    "promptEn": "scholar looks scared, wants to flee",
-    "lastPlayedAt": 1788328791080,
-    "playCount": 21
-  },
-  {
-    "id": "053bc957b4f8",
-    "source": "h3console",
-    "prompt": "道士吓得慌了神",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single twenty-one-year-old half-trained Chinese Taoist whose facial identity, tied hair, worn Taoist robe, cloth bundle, old talismans, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"frightened_panicking\". The state must be unmistakable from a single silent frame: barely controlled panic: widened eyes, a quickened silent breath, trembling tension around the mouth, and an instinct to retreat that he suppresses. Its defining motion is: He recoils slightly, glances toward an escape route, catches himself, and returns to the off-screen listener with visibly shaken composure. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: He is pushed toward the far left side of the frame in an upper-chest crop, torso recoiled away, shoulders lifted unevenly, and head pulled back from the threat. He remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish barely controlled panic: widened eyes, a quickened silent breath, trembling tension around the mouth, and an instinct to retreat that he suppresses. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into screaming, crying, fainting, confidence, attraction, broad physical retreat.\n\nHe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. His lips move rapidly in uneven fragments with sudden frozen pauses and visible trembling at each closure. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. He produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. He recoils slightly, glances toward an escape route, catches himself, and returns to the off-screen listener with visibly shaken composure. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788192913886,
-    "promptEn": "taoist panics, freaked out",
-    "lastPlayedAt": 1788330482376,
-    "playCount": 22
-  },
-  {
-    "id": "45c1136d1e05",
-    "source": "h3console",
-    "prompt": "道士慌张害怕",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single twenty-one-year-old half-trained Chinese Taoist whose facial identity, tied hair, worn Taoist robe, cloth bundle, old talismans, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"frightened_panicking\". The state must be unmistakable from a single silent frame: barely controlled panic: widened eyes, a quickened silent breath, trembling tension around the mouth, and an instinct to retreat that he suppresses. Its defining motion is: He recoils slightly, glances toward an escape route, catches himself, and returns to the off-screen listener with visibly shaken composure. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: He is pushed toward the far left side of the frame in an upper-chest crop, torso recoiled away, shoulders lifted unevenly, and head pulled back from the threat. He remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish barely controlled panic: widened eyes, a quickened silent breath, trembling tension around the mouth, and an instinct to retreat that he suppresses. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into screaming, crying, fainting, confidence, attraction, broad physical retreat.\n\nHe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. His lips move rapidly in uneven fragments with sudden frozen pauses and visible trembling at each closure. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. He produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. He recoils slightly, glances toward an escape route, catches himself, and returns to the off-screen listener with visibly shaken composure. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788192913551,
-    "promptEn": "taoist looking panicked and scared",
-    "lastPlayedAt": 1788329021239,
-    "playCount": 21
-  },
-  {
-    "id": "d169d42a67de",
-    "source": "h3console",
-    "prompt": "书生绝望地反抗",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single frail twenty-four-year-old Chinese scholar whose facial identity, tied dark hair, worn travel robe, old satchel, bookish appearance, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"desperate_defiance\". The state must be unmistakable from a single silent frame: fear held under desperate resolve: tightened brows, a firm jaw, alert eyes, and fragile courage forcing the body to remain upright. Its defining motion is: He steadies one uneven breath, lifts his chin slightly as if refusing a demand, and holds the off-screen listener's gaze before settling near the opening pose. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: He occupies the center-left from mid-chest upward, torso squared toward the listener, shoulders raised and level, with his chin held clearly higher than in the fearful state. He remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish fear held under desperate resolve: tightened brows, a firm jaw, alert eyes, and fragile courage forcing the body to remain upright. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into heroic swagger, rage, triumph, panic, smiling, combat action.\n\nHe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. His lips form firm clipped phrases with decisive closures and no hesitant trailing movement. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. He produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. He steadies one uneven breath, lifts his chin slightly as if refusing a demand, and holds the off-screen listener's gaze before settling near the opening pose. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788192913492,
-    "promptEn": "scholar defiant out of desperation",
-    "lastPlayedAt": 1788330716569,
-    "playCount": 22
-  },
-  {
-    "id": "2abd2793fae9",
-    "source": "h3console",
-    "prompt": "书生走投无路还在反抗",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single frail twenty-four-year-old Chinese scholar whose facial identity, tied dark hair, worn travel robe, old satchel, bookish appearance, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"desperate_defiance\". The state must be unmistakable from a single silent frame: fear held under desperate resolve: tightened brows, a firm jaw, alert eyes, and fragile courage forcing the body to remain upright. Its defining motion is: He steadies one uneven breath, lifts his chin slightly as if refusing a demand, and holds the off-screen listener's gaze before settling near the opening pose. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: He occupies the center-left from mid-chest upward, torso squared toward the listener, shoulders raised and level, with his chin held clearly higher than in the fearful state. He remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish fear held under desperate resolve: tightened brows, a firm jaw, alert eyes, and fragile courage forcing the body to remain upright. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into heroic swagger, rage, triumph, panic, smiling, combat action.\n\nHe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. His lips form firm clipped phrases with decisive closures and no hesitant trailing movement. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. He produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. He steadies one uneven breath, lifts his chin slightly as if refusing a demand, and holds the off-screen listener's gaze before settling near the opening pose. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788192913244,
-    "promptEn": "scholar cornered but still defiant",
-    "lastPlayedAt": 1788327408421,
-    "playCount": 21
-  },
-  {
-    "id": "bbf7efb467e0",
-    "source": "h3console",
-    "prompt": "道士被迷得犹豫不决",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single twenty-one-year-old half-trained Chinese Taoist whose facial identity, tied hair, worn Taoist robe, cloth bundle, old talismans, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"charmed_hesitating\". The state must be unmistakable from a single silent frame: conflicted fascination: softened eyes alternate with a guarded brow, the mouth hesitates between phrases, and suspicion repeatedly interrupts attraction. Its defining motion is: He leans forward almost imperceptibly, stops himself, looks down for a beat, then returns his gaze with an uneasy mixture of caution and attraction. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: He occupies the center-right in a closer upper-chest crop, torso leaning toward the listener while his head remains slightly turned away in visible conflict. He remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish conflicted fascination: softened eyes alternate with a guarded brow, the mouth hesitates between phrases, and suspicion repeatedly interrupts attraction. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into complete surrender, romance, sleepiness, comic embarrassment, panic, confident rejection.\n\nHe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. His lips begin softly, stop halfway through several movements, press closed, and restart with hesitant slower articulation. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. He produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. He leans forward almost imperceptibly, stops himself, looks down for a beat, then returns his gaze with an uneasy mixture of caution and attraction. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788192912912,
-    "promptEn": "taoist charmed, hesitating",
-    "lastPlayedAt": 1788331160933,
-    "playCount": 22
-  },
-  {
-    "id": "f8da9934c823",
-    "source": "h3console",
-    "prompt": "书生被迷得恍恍惚惚",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single frail twenty-four-year-old Chinese scholar whose facial identity, tied dark hair, worn travel robe, old satchel, bookish appearance, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"charmed_dazed\". The state must be unmistakable from a single silent frame: subtle enchanted distraction: softened focus, slower blinking, loosened facial tension, and a faint involuntary pull toward the off-screen listener. Its defining motion is: His eyes linger a moment too long, his head inclines slightly toward the off-screen listener, then he blinks and almost regains his guarded posture. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: He occupies the center-right of the frame in a slightly closer upper-chest crop, torso leaning subtly toward the listener and head inclined in the same direction. He remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish subtle enchanted distraction: softened focus, slower blinking, loosened facial tension, and a faint involuntary pull toward the off-screen listener. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into sleep, intoxication, broad romance, comic confusion, vacant unconsciousness, overt supernatural effects.\n\nHe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. His lips move slowly and softly with delayed starts, longer pauses, and slightly prolonged final mouth closures. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. He produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. His eyes linger a moment too long, his head inclines slightly toward the off-screen listener, then he blinks and almost regains his guarded posture. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788192912819,
-    "promptEn": "scholar dazed, charmed",
-    "lastPlayedAt": 1788330684533,
-    "playCount": 22
-  },
-  {
-    "id": "feb96d5be50f",
-    "source": "h3console",
-    "prompt": "白衣女子被吓到，有点害羞",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single pale-skinned young Chinese woman whose facial identity, high dark hair bun, sheer white traditional robe, delicate features, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"startled_shy\". The state must be unmistakable from a single silent frame: clear startled shyness: both shoulders draw inward, her head turns fifteen degrees away, her eyes remain visibly widened beneath a raised guarded brow, and her lips hold uncertain tension. Its defining motion is: She keeps the inward posture, lowers her gaze for a full beat, checks the off-screen intruder from the corner of her eyes, and returns without straightening her shoulders. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: She occupies the right third of the frame from mid-chest upward, torso turned thirty degrees away from the intruder, shoulders drawn inward, and head angled toward the nearest exit. She remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish clear startled shyness: both shoulders draw inward, her head turns fifteen degrees away, her eyes remain visibly widened beneath a raised guarded brow, and her lips hold uncertain tension. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into seduction, warmth, anger, supernatural menace, crying, theatrical fear.\n\nShe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. Her lips form brief hesitant movements with long closed-mouth pauses, while the corners remain tense and never form a smile. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. She produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. She keeps the inward posture, lowers her gaze for a full beat, checks the off-screen intruder from the corner of her eyes, and returns without straightening her shoulders. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788191432903,
-    "promptEn": "woman in white startled, a bit shy",
-    "lastPlayedAt": 1788329539599,
-    "playCount": 21
-  },
-  {
-    "id": "14f482bdae7f",
-    "source": "h3console",
-    "prompt": "白衣女子亲密地魅惑对方",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single pale-skinned young Chinese woman whose facial identity, high dark hair bun, sheer white traditional robe, delicate features, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"intimate_charming\". The state must be unmistakable from a single silent frame: clear intimate allure: her head remains tilted ten degrees toward the listener, her eyelids stay visibly lowered, a restrained knowing smile remains at one corner of her mouth, and her gaze is unwaveringly personal. Its defining motion is: She preserves the tilted head and lowered eyelids, leans slightly closer for one beat, lets the gaze linger, and eases back without losing the knowing half-smile. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: She occupies the center-left in a noticeably closer upper-chest crop, torso angled toward the listener, head tilted ten degrees, and face slightly nearer the camera. She remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish clear intimate allure: her head remains tilted ten degrees toward the listener, her eyelids stay visibly lowered, a restrained knowing smile remains at one corner of her mouth, and her gaze is unwaveringly personal. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into broad flirtation, laughter, explicit sexuality, supernatural glow, anger, vulnerability.\n\nShe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. Her lips move slowly with soft precise articulation, longer lingering closures, and the knowing half-smile returning between movements. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. She produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. She preserves the tilted head and lowered eyelids, leans slightly closer for one beat, lets the gaze linger, and eases back without losing the knowing half-smile. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788191432362,
-    "promptEn": "woman in white being intimate and alluring",
-    "lastPlayedAt": 1788331305059,
-    "playCount": 22
-  },
-  {
-    "id": "362b15ab1811",
-    "source": "h3console",
-    "prompt": "白衣女子温柔安慰人",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single pale-skinned young Chinese woman whose facial identity, high dark hair bun, sheer white traditional robe, delicate features, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"gentle_comforting\". The state must be unmistakable from a single silent frame: clear reassuring warmth: her shoulders are open and relaxed, her head stays upright, her brow is smooth, and her soft direct gaze remains steady without coyness. Its defining motion is: She maintains open posture, gives one slow calming nod, briefly acknowledges the doorway with her eyes, and returns to stable direct eye contact. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: She stands centered from mid-chest upward, torso open toward the listener, shoulders relaxed and level, head upright, and face held at a neutral conversational distance. She remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish clear reassuring warmth: her shoulders are open and relaxed, her head stays upright, her brow is smooth, and her soft direct gaze remains steady without coyness. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into overt seduction, fear, suspicion, anger, broad smiling, maternal exaggeration.\n\nShe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. Her lips move in a slow even cadence with gentle rounded articulation and calm closed-mouth pauses. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. She produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. She maintains open posture, gives one slow calming nod, briefly acknowledges the doorway with her eyes, and returns to stable direct eye contact. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788191431874,
-    "promptEn": "woman in white gently comforting someone",
-    "lastPlayedAt": 1788326391548,
-    "playCount": 21
-  },
-  {
-    "id": "a2c515055269",
-    "source": "h3console",
-    "prompt": "白衣女子害怕地搪塞",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single pale-skinned young Chinese woman whose facial identity, high dark hair bun, sheer white traditional robe, delicate features, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"fearful_deflecting\". The state must be unmistakable from a single silent frame: clear restless false fear: her shoulders remain uneven and slightly hunched, her brow is asymmetrically tightened, her worried eyes repeatedly avoid direct contact, and her lips press together between bursts of speech. Its defining motion is: She makes two quick sideward eye movements toward the threatened evidence, briefly presses her lips closed, then looks back with deliberately vulnerable widened eyes while her shoulders remain uneven. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: She occupies the left third from mid-chest upward, torso twisted fifteen degrees toward the threatened evidence, shoulders visibly uneven, and head caught between the evidence and listener. She remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish clear restless false fear: her shoulders remain uneven and slightly hunched, her brow is asymmetrically tightened, her worried eyes repeatedly avoid direct contact, and her lips press together between bursts of speech. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into open rage, overt threat, genuine panic, seduction, smiling, physical obstruction.\n\nShe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. Her lips move in short irregular bursts, press firmly closed after each burst, and restart only after a quick sideward glance. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. She produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. She makes two quick sideward eye movements toward the threatened evidence, briefly presses her lips closed, then looks back with deliberately vulnerable widened eyes while her shoulders remain uneven. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788191431505,
-    "promptEn": "woman in white scared, dodging the question",
-    "lastPlayedAt": 1788330838667,
-    "playCount": 22
-  },
-  {
-    "id": "667978d52305",
-    "source": "h3console",
-    "prompt": "白衣女子冷冷地威胁",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single pale-skinned young Chinese woman whose facial identity, high dark hair bun, sheer white traditional robe, delicate features, and period styling come from <Picture 2>. <Picture 2> is the approved state reference frame: preserve its facial identity, clothing, hairstyle, body proportions, expression, gaze, head angle, posture, and state-specific composition.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"cold_threatening\". The state must be unmistakable from a single silent frame: clear predatory menace: her shoulders are level and rigid, her chin stays lowered eight degrees, her jaw is firmly set, her eyes hold direct unbroken contact, and she does not blink or show warmth. Its defining motion is: She remains almost motionless, keeps unwavering direct eye contact, lowers her chin once with deliberate precision, and returns to the same rigid threatening pose without any sideward glance. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> according to this state-specific composition: She stands exactly centered in a tighter upper-chest crop, torso square to the target, shoulders rigid and level, head perfectly vertical, and face symmetrically framed by the screens. She remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish clear predatory menace: her shoulders are level and rigid, her chin stays lowered eight degrees, her jaw is firmly set, her eyes hold direct unbroken contact, and she does not blink or show warmth. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into shouting, snarling, monster transformation, visible attack, theatrical rage, smiling.\n\nShe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. Her lips make minimal precise movements at a slow controlled cadence, closing into a firm straight line between phrases without any smile or tremble. The movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. She produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. She remains almost motionless, keeps unwavering direct eye contact, lowers her chin once with deliberate precision, and returns to the same rigid threatening pose without any sideward glance. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788191431016,
-    "promptEn": "woman in white coldly threatening",
-    "lastPlayedAt": 1788330486379,
-    "playCount": 22
-  },
-  {
-    "id": "7f5213748d87",
-    "source": "h3console",
-    "prompt": "白衣女子害怕地搪塞",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single pale-skinned young Chinese woman whose facial identity, high dark hair bun, sheer white traditional robe, delicate features, and period styling come from <Picture 2>. <Picture 2> is used only as an identity, clothing, hairstyle, body-proportion, and accessory reference; its source composition, borders, labels, empty background, and design-board layout never appear in the target video.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"fearful_deflecting\". The state must be unmistakable from a single silent frame: clear restless false fear: her shoulders remain uneven and slightly hunched, her brow is asymmetrically tightened, her worried eyes repeatedly avoid direct contact, and her lips press together between bursts of speech. Its defining motion is: She makes two quick sideward eye movements toward the threatened evidence, briefly presses her lips closed, then looks back with deliberately vulnerable widened eyes while her shoulders remain uneven. It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> from the upper chest upward. She stands nearly front-facing with her torso angled slightly toward an off-screen conversation partner. She remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nBefore any mouth movement begins, the face, gaze, head angle, jaw, and shoulders clearly establish clear restless false fear: her shoulders remain uneven and slightly hunched, her brow is asymmetrically tightened, her worried eyes repeatedly avoid direct contact, and her lips press together between bursts of speech. This visual signature remains continuously readable throughout the shot and never relaxes into a generic neutral speaking face. It never shifts into open rage, overt threat, genuine panic, seduction, smiling, physical obstruction.\n\nShe begins natural speech-related mouth movement while preserving the defining expression, gaze direction, head angle, jaw tension, and shoulder posture. She opens and closes her lips with varied articulation and pauses, but the movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. She produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. She makes two quick sideward eye movements toward the threatened evidence, briefly presses her lips closed, then looks back with deliberately vulnerable widened eyes while her shoulders remain uneven. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788188423583,
-    "promptEn": "woman in white scared, dodging",
-    "lastPlayedAt": 1788327056113,
-    "playCount": 21
-  },
-  {
-    "id": "1597db4be561",
-    "source": "h3console",
-    "prompt": "白衣女子温柔安慰",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single pale-skinned young Chinese woman whose facial identity, high dark hair bun, sheer white traditional robe, delicate features, and period styling come from <Picture 2>. <Picture 2> is used only as an identity, clothing, hairstyle, body-proportion, and accessory reference; its source composition, borders, labels, empty background, and design-board layout never appear in the target video.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"gentle_comforting\". It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. The performance is controlled gentleness and reassuring patience: soft attentive eyes, relaxed lips, a composed brow, and measured warmth that remains slightly guarded. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> from the upper chest upward. She stands nearly front-facing with her torso angled slightly toward an off-screen conversation partner. She remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nAt the opening, the performance conveys controlled gentleness and reassuring patience: soft attentive eyes, relaxed lips, a composed brow, and measured warmth that remains slightly guarded. The emotion remains believable and controlled. It never shifts into overt seduction, fear, suspicion, anger, broad smiling, maternal exaggeration.\n\nShe begins natural speech-related mouth movement at a measured conversational pace. She opens and closes her lips with varied, moderate articulation and brief pauses, but the movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. She produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. She gives a small calming nod, briefly glances toward the door as if acknowledging danger, then returns her full attention to the off-screen listener. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788187365789,
-    "lastPlayedAt": 1788328905176,
-    "promptEn": "woman in white being gentle and comforting",
-    "playCount": 21
-  },
-  {
-    "id": "88d71d85058f",
-    "source": "h3console",
-    "prompt": "白衣女子躲闪着回避问题",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single pale-skinned young Chinese woman whose facial identity, high dark hair bun, sheer white traditional robe, delicate features, and period styling come from <Picture 2>. <Picture 2> is used only as an identity, clothing, hairstyle, body-proportion, and accessory reference; its source composition, borders, labels, empty background, and design-board layout never appear in the target video.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"fearful_deflecting\". It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. The performance is manufactured fear concealing calculation: worried eyes, a tightened brow, slightly uneven composure, and a quick assessment of whether the deception is working. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> from the upper chest upward. She stands nearly front-facing with her torso angled slightly toward an off-screen conversation partner. She remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nAt the opening, the performance conveys manufactured fear concealing calculation: worried eyes, a tightened brow, slightly uneven composure, and a quick assessment of whether the deception is working. The emotion remains believable and controlled. It never shifts into open rage, overt threat, genuine panic, seduction, smiling, physical obstruction.\n\nShe begins natural speech-related mouth movement at a measured conversational pace. She opens and closes her lips with varied, moderate articulation and brief pauses, but the movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. She produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. Her eyes dart once toward the threatened evidence outside the frame, she catches the reaction, then turns back with a carefully vulnerable expression. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788186218619,
-    "lastPlayedAt": 1788329921918,
-    "promptEn": "woman in white evading, scared",
-    "playCount": 22
-  },
-  {
-    "id": "6cb4eb75ee09",
-    "source": "h3console",
-    "prompt": "白衣女子冷冷地发出威胁",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single pale-skinned young Chinese woman whose facial identity, high dark hair bun, sheer white traditional robe, delicate features, and period styling come from <Picture 2>. <Picture 2> is used only as an identity, clothing, hairstyle, body-proportion, and accessory reference; its source composition, borders, labels, empty background, and design-board layout never appear in the target video.\n<Subject 2> is the old Chinese room whose spatial layout, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber lighting, and deep surrounding shadows come from <Picture 1>. <Picture 1> is used only as the environment, composition, lighting, and period-style reference.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for <Subject 1> in the finite performance state \"cold_threatening\". It shows only <Subject 1> in a stable medium close-up inside <Subject 2>. The performance is controlled predatory menace: still eyes, lowered warmth, a firm mouth, minimal blinking, and absolute attention on the off-screen target. Natural speech-related mouth movement does not correspond to any sentence, language, or fixed phoneme sequence. The ending returns close to the opening posture so the clip can repeat under variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, age, clothing, hairstyle, body proportions, accessories, and period styling from <Picture 2> remain consistent throughout the video.\n<Subject 2> (appears in [Shot 1]): partially_preserved - the room identity, carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows from <Picture 1> remain recognizable behind the closer character framing; no person from the source image is copied.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained acting, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> from the upper chest upward. She stands nearly front-facing with her torso angled slightly toward an off-screen conversation partner. She remains the same person defined by <Picture 2>. Behind the character is <Subject 2>, preserving the environment established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in shadow. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nAt the opening, the performance conveys controlled predatory menace: still eyes, lowered warmth, a firm mouth, minimal blinking, and absolute attention on the off-screen target. The emotion remains believable and controlled. It never shifts into shouting, snarling, monster transformation, visible attack, theatrical rage, smiling.\n\nShe begins natural speech-related mouth movement at a measured conversational pace. She opens and closes her lips with varied, moderate articulation and brief pauses, but the movement does not match any identifiable sentence, language, or fixed phoneme sequence. The mouth never stretches into exaggerated shapes. She produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. Her head becomes unnaturally still, her gaze hardens without widening, and her chin lowers slightly before she returns to the opening threatening pose. Hands remain outside the frame, and no semantic action binds the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The intended emotional performance remains legible while the gaze, head angle, shoulder height, facial tension, lighting, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nN/A\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788185090875,
-    "promptEn": "woman in white cold and threatening",
-    "lastPlayedAt": 1788327712623,
-    "playCount": 21
-  },
-  {
-    "id": "a46f1daa9ee0",
-    "source": "h3console",
-    "prompt": "巨大战舰从星空中驶过，舰桥突然剧烈震动",
-    "fullPrompt": "[Shot 1] Photorealistic cinematic science-fiction teaser footage in a cold blue-black palette, lit only by starlight. A vast deep-space star field fills the frame, a near-black emptiness across the left third, a dense band of bright stars in the upper right. A colossal dark-gray warship silhouette glides in from the left edge toward mid-frame, its ridged hull catching faint cold starlight on one flank while the opposite side sinks into darkness. The camera trucks slowly right with the hull.\n\n[Shot 2] At 00:02.500, the shot cuts to the warship's dim bridge interior. Cold white starlight pours through a multi-pane forward window, consoles glowing with small amber instrument lights. A heavy vibration builds: bulkheads, consoles and a backlit human silhouette bracing one hand on a console all shudder together in the same direction, while the camera holds static and amber light spots sweep the walls.\n\n[Shot 3] At 00:05.500, the shot cuts to the curved limb of a cold gray-blue planet entering from the right, its day-night boundary sharply defined, lit arc cold white, dark side merging into black space. The camera pushes in slowly.\n\n[Shot 4] At 00:06.750, the shot cuts to a distant group of at least three warships of the same hull family, spread unevenly with slight depth separation. The camera pulls out slowly as the dark hulls shrink against the star field, unmarked except faint blue-white engine points.\n\n[Shot 5] At 00:08.000, the shot cuts to a close pass along the lead warship's hull. The camera tracks along the ship's longitudinal spine, revealing broad polygonal armor plates with deep seams, parallel channels down both flanks, a ring reinforcement collar near the bow, and one faint amber cabin glow as a warm accent.\n\n[Shot 6] At 00:09.250, the shot cuts to the fleet inside a warp transition. Stars stretch into radiating cold white and pale-cyan light lines while space visibly distorts, and the camera pushes in through the streaking field. At 0.6 seconds into this shot, the frame's brightness snaps up to a near-overexposed white flash for an instant and immediately falls back, leaving a few frames of lens-flare streaks as the fleet jumps away.\n\n[Shot 7] At 00:11.125, the shot cuts to total blackness. The camera holds static.\n\n[Shot 8] At 00:11.500, the shot cuts back to the darkest region of the star field. The camera holds static as the four Chinese characters \"星渊纪元\" emerge gradually out of the darkness, their brightness rising from near-black instead of cutting in abruptly. The characters are set with wide letter spacing of at least half a character width, rendered as a brushed dark metal gray rather than pure white. Partway through the shot a narrow band of starlight sweeps across the letter surfaces, brightening each character it passes and leaving a decaying afterimage trail; during the sweep the characters visibly spread apart, then settle. Afterward they keep a restrained low-intensity glow, with a faint edge glow that stays clear of the neighboring characters and never swallows the outlines.\n\noverall_soundscape: A continuous deep sub-bass drone underscores the whole piece. Structural rumble and stress groans build through the bridge vibration. Short whoosh impacts punctuate each hard cut, and a deep low-frequency drop slams exactly with the white flash of the jump. The sound briefly thins to near-silence over the black frame before the music's final hit lands as the title settles.\n\nnon_diegetic_music: Rhythmic percussion enters over the low drone at the hard-cut sequence, lands repeated accent hits through the jump, and reaches its final peak as the title locks into place. No melodic vocals appear at any point.\n",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 14.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1788181569491,
-    "promptEn": "a massive warship glides through the stars, the bridge suddenly shakes hard",
-    "lastPlayedAt": 1788328747041,
-    "playCount": 21
-  },
-  {
-    "id": "44bf5b969e94",
-    "source": "h3console",
-    "prompt": "古装女子在老宅里说话，神情温柔又警惕",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single pale-skinned young woman whose facial identity, high dark hair bun, sheer white traditional robe, delicate features, and period styling come from <Picture 2>. <Picture 2> is used only as an identity, clothing, hairstyle, and accessory reference; its multi-view character-sheet layout, auxiliary views, labels, borders, and design-board composition never appear in the target video.\n<Picture 1> establishes the old Chinese room used as the target video's environment: five carved wooden screens, the round stone table and oil lamp, the daybed along the right wall, hanging scrolls, warm amber light, and deep surrounding shadows.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for the finite expression state \"gentle_but_alert\". It shows only <Subject 1> in a stable medium close-up inside the room established by <Picture 1>. Her expression remains gentle while retaining subtle caution, and her natural speech-related mouth movement does not correspond to any sentence. The performance finishes close to its opening posture so the clip can repeat during variable-length TTS playback.\n\nretention_analysis:\n<Picture 1> ([Shot 1] environment): partially_preserved - the carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows define one coherent room behind the closer character framing; no person from the source image is copied.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, high dark hair bun, sheer white traditional robe, pale complexion, age, body proportions, hairstyle, and period styling from <Picture 2> remain consistent throughout the video.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained performance, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> from the upper chest upward. She stands nearly front-facing with her torso angled slightly toward an off-screen conversation partner. Her facial identity, high dark hair bun, sheer white robe collar, layered fabric, and period appearance follow <Picture 2>. Behind her is the same room established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in the shadows. The environment is a single physical room rather than a copied flat backdrop. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nDuring the opening moment, <Subject 1> holds a composed resting posture and looks toward the off-screen conversation partner. Her expression is outwardly gentle but alert: relaxed lips, slightly attentive eyes, a faintly guarded brow, and minimal tension in the jaw. The emotion remains believable and controlled, without smiling broadly, theatrical fear, anger, crying, or sudden body movement.\n\nShe begins natural speech-related mouth movement at a calm conversational pace. Her lips open and close with varied, moderate articulation and brief pauses, but the movement does not match any identifiable sentence, language, or fixed phoneme sequence. Her mouth never stretches into exaggerated shapes. She produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. While continuing the generic speaking motion, her eyes briefly study the off-screen conversation partner, then glance toward the room entrance before returning. Her chin lowers slightly and settles. Her shoulders remain relaxed, and her hands stay outside the frame. She performs no semantic action that would bind the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The gentle but guarded expression remains visible, while the gaze, head angle, shoulder height, facial tension, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nComplete silence. No dialogue, voice, whisper, breathing, mouth sound, footsteps, clothing movement, room tone, environmental sound, or sound effect is audible.\n\nnon_diegetic_music:\nNone. No music is audible.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788162094002,
-    "promptEn": "woman in old chinese robe talking, gentle but wary, in an old room",
-    "lastPlayedAt": 1788330508397,
-    "playCount": 22
-  },
-  {
-    "id": "e17cfe57f56a",
-    "source": "h3console",
-    "prompt": "古装女子在老宅里说话，神情温柔又警惕",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single pale-skinned young woman whose facial identity, high dark hair bun, sheer white traditional robe, delicate features, and period styling come from <Picture 2>. <Picture 2> is used only as an identity, clothing, hairstyle, and accessory reference; its multi-view character-sheet layout, auxiliary views, labels, borders, and design-board composition never appear in the target video.\n<Picture 1> establishes the old Chinese room used as the target video's environment: five carved wooden screens, the round stone table and oil lamp, the daybed along the right wall, hanging scrolls, warm amber light, and deep surrounding shadows.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for the finite expression state \"gentle_but_alert\". It shows only <Subject 1> in a stable medium close-up inside the room established by <Picture 1>. Her expression remains gentle while retaining subtle caution, and her natural speech-related mouth movement does not correspond to any sentence. The performance finishes close to its opening posture so the clip can repeat during variable-length TTS playback.\n\nretention_analysis:\n<Picture 1> ([Shot 1] environment): partially_preserved - the carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows define one coherent room behind the closer character framing; no person from the source image is copied.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, high dark hair bun, sheer white traditional robe, pale complexion, age, body proportions, hairstyle, and period styling from <Picture 2> remain consistent throughout the video.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained performance, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> from the upper chest upward. She stands nearly front-facing with her torso angled slightly toward an off-screen conversation partner. Her facial identity, high dark hair bun, sheer white robe collar, layered fabric, and period appearance follow <Picture 2>. Behind her is the same room established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in the shadows. The environment is a single physical room rather than a copied flat backdrop. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nDuring the opening moment, <Subject 1> holds a composed resting posture and looks toward the off-screen conversation partner. Her expression is outwardly gentle but alert: relaxed lips, slightly attentive eyes, a faintly guarded brow, and minimal tension in the jaw. The emotion remains believable and controlled, without smiling broadly, theatrical fear, anger, crying, or sudden body movement.\n\nShe begins natural speech-related mouth movement at a calm conversational pace. Her lips open and close with varied, moderate articulation and brief pauses, but the movement does not match any identifiable sentence, language, or fixed phoneme sequence. Her mouth never stretches into exaggerated shapes. She produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. While continuing the generic speaking motion, her eyes briefly study the off-screen conversation partner, then glance toward the room entrance before returning. Her chin lowers slightly and settles. Her shoulders remain relaxed, and her hands stay outside the frame. She performs no semantic action that would bind the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The gentle but guarded expression remains visible, while the gaze, head angle, shoulder height, facial tension, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nComplete silence. No dialogue, voice, whisper, breathing, mouth sound, footsteps, clothing movement, room tone, environmental sound, or sound effect is audible.\n\nnon_diegetic_music:\nNone. No music is audible.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788160917846,
-    "promptEn": "woman in old chinese robe talking, gentle but wary, in an old room",
-    "lastPlayedAt": 1788329937932,
-    "playCount": 22
-  },
-  {
-    "id": "24e49e5ae069",
-    "source": "h3console",
-    "prompt": "捕快在老宅里说话，神情紧张害怕",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single young male constable whose facial identity, black headscarf, dark indigo robe, leather waist belt, and period styling come from <Picture 2>. <Picture 2> is used only as an identity, clothing, hairstyle, and accessory reference; its multi-view character-sheet layout, auxiliary views, labels, borders, and design-board composition never appear in the target video.\n<Picture 1> establishes the old Chinese room used as the target video's environment: five carved wooden screens, the round stone table and oil lamp, the daybed along the right wall, hanging scrolls, warm amber light, and deep surrounding shadows.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for the finite expression state \"afraid\". It shows only <Subject 1> in a stable medium close-up inside the room established by <Picture 1>. He displays restrained fear while making natural speech-related mouth movements that do not correspond to any sentence. The performance finishes close to its opening posture so the clip can repeat during variable-length TTS playback.\n\nretention_analysis:\n<Picture 1> ([Shot 1] environment): partially_preserved - the carved wooden screens, round stone table, oil lamp, daybed, hanging scrolls, warm amber light, and deep shadows define one coherent room behind the closer character framing; no person from the source image is copied.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, black headscarf, dark indigo constable robe, leather belt, age, body proportions, hairstyle, and period styling from <Picture 2> remain consistent throughout the video.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-horror style with realistic skin texture, restrained performance, warm oil-lamp illumination, deep brown-black shadows, and shallow depth of field. It is one continuous normal cinematic shot containing one character. It never becomes a character sheet, collage, split screen, turnaround view, portrait grid, close-up inset, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> from the upper chest upward. He stands nearly front-facing with his torso angled slightly toward an off-screen conversation partner. His facial identity, black headscarf, dark indigo robe collar, leather shoulder details, and period appearance follow <Picture 2>. Behind him is the same room established by <Picture 1>: recognizable carved wooden screens occupy the softly blurred middle ground, while the round stone table and its single oil lamp create warm side light; the daybed and hanging scrolls remain subdued in the shadows. The environment is a single physical room rather than a copied flat backdrop. No additional person appears. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nDuring the opening moment, <Subject 1> holds a guarded resting posture and looks toward the off-screen conversation partner. Restrained fear develops through slightly widened eyes, gently tightened eyebrows, shallow breathing, mild tension around the jaw, and a small swallow. The emotion remains believable and controlled, without theatrical grimacing, screaming, crying, or panic-driven body movement.\n\nHe begins natural speech-related mouth movement at a measured conversational pace. His lips open and close with varied, moderate articulation and brief pauses, but the movement does not match any identifiable sentence, language, or fixed phoneme sequence. His mouth never stretches into exaggerated shapes. He produces no audible dialogue, vocalization, whisper, breathing sound, or mouth sound. While continuing the generic speaking motion, his eyes briefly glance toward the edge of the frame as if checking for a nearby threat, then return to the off-screen conversation partner. His head draws back by only a few centimeters and settles. His shoulders rise subtly with one silent breath and lower again. His hands remain outside the frame, and he performs no semantic action that would bind the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The frightened alertness remains visible, while the gaze, head angle, shoulder height, facial tension, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape:\nComplete silence. No dialogue, voice, whisper, breathing, mouth sound, footsteps, clothing movement, room tone, environmental sound, or sound effect is audible.\n\nnon_diegetic_music:\nNone. No music is audible.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788159367556,
-    "promptEn": "constable talking nervously in an old room, looking scared",
-    "lastPlayedAt": 1788328967208,
-    "playCount": 21
-  },
-  {
-    "id": "a6d767413426",
-    "source": "h3console",
-    "prompt": "捕快紧张害怕地说着话",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single young male constable whose facial identity, black headscarf, dark indigo robe, leather waist belt, and period styling come from <Picture 1>. <Picture 1> is used only as an identity, clothing, hairstyle, and accessory reference; its multi-view character-sheet layout is not reproduced in the target video.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for the finite expression state \"afraid\". It shows only <Subject 1> in a stable medium close-up, displaying restrained fear while making natural speech-related mouth movements that do not correspond to any specific sentence. The performance finishes close to its opening posture so the clip can repeat during variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, black headscarf, dark indigo constable robe, leather belt, age, body proportions, hairstyle, and period styling from <Picture 1> remain consistent throughout the video.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-mystery style with realistic skin texture, restrained performance, soft warm key light, dim brown-black surroundings, and shallow depth of field. The output is a normal cinematic frame containing one character, never a character sheet, collage, split screen, turnaround view, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> from the upper chest upward. He stands nearly front-facing with his torso angled slightly toward an off-screen conversation partner. His facial identity, black headscarf, dark indigo robe collar, leather shoulder details, and period appearance follow <Picture 1>. The background is a softly blurred, dim old Chinese interior with no readable signs, no additional people, and no distinct story event. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nDuring the opening moment, <Subject 1> holds a guarded resting posture and looks toward the off-screen conversation partner. Restrained fear develops through slightly widened eyes, gently tightened eyebrows, shallow breathing, mild tension around the jaw, and a small swallow. The emotion remains believable and controlled, without theatrical grimacing, screaming, crying, or panic-driven body movement.\n\nHe begins natural speech-related mouth movement at a measured conversational pace. His lips open and close with varied, moderate articulation and brief pauses, but the movement does not match any identifiable sentence, language, or fixed phoneme sequence. His mouth never stretches into exaggerated shapes. While continuing the generic speaking motion, his eyes briefly glance toward the edge of the frame as if checking for a nearby threat, then return to the off-screen conversation partner. His head draws back by only a few centimeters and settles. His shoulders rise subtly with one quiet breath and lower again. His hands remain outside the frame, and he performs no semantic action that would bind the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The frightened alertness remains visible, while the gaze, head angle, shoulder height, facial tension, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape: N/A\n\nnon_diegetic_music: N/A",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788157322611,
-    "promptEn": "constable talking, looking scared",
-    "lastPlayedAt": 1788327456447,
-    "playCount": 21
-  },
-  {
-    "id": "7977eeeabe8b",
-    "source": "h3console",
-    "prompt": "捕快慌张闯进屋抵住门，跟屏风后的白衣女子说话",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the woman in <Picture 2>: a mysterious woman in a plain white ancient-Chinese gown, with a delicate but unnaturally pale face, long jet-black hair worn loose with no ornaments, standing motionless with her hem reaching the floor.\n<Subject 2> is the man in <Picture 3>: an ancient Chinese constable in a dark ink-black long robe reaching his ankles, with a black headscarf tied at the back of his head, a resolute square face with strong brows, a sheathed saber at his left hip, a round wooden waist tag reading \"捕\" on his right hip, and black short boots.\n<Picture 1> is the scene reference for the dim old room and the composition anchor for [Shot 1]: five carved openwork screen panels standing in a row at the center, a round stone-top table carrying a single cold unlit candle with its bare wick exposed and no flame or glow anywhere on it — a wooden luohan bed on the right side, a paper wall scroll on the wall, and a heavy wooden door on the left.\n\nsummary:\n[reference generation] A panicked ancient constable (<Subject 2>) bursts through the door of a dim old Chinese room at night, braces the door shut, and speaks a single line of dialogue to the pale woman in white (<Subject 1>) who steps out from behind the five-panel carved screen; the only lit object in the room is one small oil lamp, the candle on the stone table stays unlit, no pursuer is ever shown or heard, and no music plays.\n\nretention_analysis:\n<Subject 1>: partially_preserved - the white gown, pale face and long black hair from <Picture 2> are retained; her walk from behind the screen, her head turn and her stillness in [Shot 4] and [Shot 5] are newly generated, and she is absent from [Shot 1] through [Shot 3].\n<Subject 2>: partially_preserved - the dark robe, black headscarf, saber and waist tag from <Picture 3> are retained; his panicked entrance, bracing of the door and spoken line are newly generated.\n<Picture 1>: partially_preserved - the room layout, the five-panel screen, the stone table with its cold unlit candle, the luohan bed and the wall scroll are retained as the setting; the lit oil lamp beside the luohan bed, its flame, the cast shadows and the night atmosphere are newly generated.\n\ndetailed_description:\nLive-action cinematic Chinese period horror in a dim old Chinese room at night. The room contains exactly one working light source: a small squat bronze oil lamp with a low steady open flame on the low cabinet beside the luohan bed at the mid-ground right; every other object is unlit, and all warm light comes only from that flame. [Shot 1] In the room shown in <Picture 1>, the heavy wooden door on the left wall is flung open and the constable (<Subject 2>) stumbles in, catching his balance while the camera tracks forward from the doorway behind him, his dark robe swallowing what little light reaches him. The five carved screen panels stand in the deep center of the frame, and the oil lamp's low flame is pressed flat for a moment by the draft he brings in. [Shot 2] At 00:01.200, the camera holds a static medium shot as the constable braces the door shut with both hands, his back to the room, the oil lamp flame ducking low again under the draft and slowly recovering. [Shot 3] At 00:02.000, the camera pulls back from the constable to reveal the whole room: the five openwork carved panels throw striped shadows onto the wall behind them, the stripes swaying with the lamp's flicker, and the round stone-top table comes into view with the candle on it completely dead — a white wax pillar in a bronze holder, bare wick exposed and cold, no flame, no halo of light around its tip, no pool of warm light on the tabletop — while the luohan bed on the right and the paper wall scroll also come into view, the constable a small foreground figure at the left. [Shot 4] At 00:03.200, the camera pushes in slowly toward the screen as the woman (<Subject 1>) steps out from behind the panels, her hem dragging and only her upper body moving, then stops in the lamplight and turns her head toward the constable in the left depth of the same frame; her pale face reads colder than the warm lamplight around it, no part of her is visible in any earlier shot, and the oil lamp stays inside the frame throughout, still the only lit object. [Shot 5] At 00:05.400, the camera trucks slowly rightward from the constable toward the woman: bracing the door without retreating, he lowers his voice and says <d>[Chinese] 对不起，我不是有意闯进来的，外面有东西在追我。</d> while the oil lamp flame trembles faintly with his voice; her white gown is the brightest shape in the room, and after the line both figures stand motionless facing each other across the screen panels, the shot ending on the two of them, the candle on the stone table still cold and dead. His saber stays sheathed for the whole clip, nothing and no one is ever shown or heard outside the closed door, and no second light source — no burning candle, no standing lantern — ever appears in the room.\n\noverall_soundscape:\nA heavy door slam and a muffled wooden groan open the clip under the constable's (<Subject 2>) rapid ragged panting, and a draft of night wind seeps in from outside the closed door. The single oil lamp contributes a faint wick crackle that continues under the whole clip. Soft cloth rustle and nearly silent footsteps mark the woman (<Subject 1>) stepping out from behind the screen. After the constable (<Subject 2>) finishes his line, the room falls into dead silence with only the faint flicker of the lamp flame.\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788157177505,
-    "promptEn": "constable bursts in and blocks the door, talks to a pale woman behind the screen",
-    "lastPlayedAt": 1788328452807,
-    "playCount": 21
-  },
-  {
-    "id": "a9c4ba5e453b",
-    "source": "h3console",
-    "prompt": "捕快慌张闯进屋抵住门，跟屏风后的白衣女子说话",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the woman in <Picture 2>: a mysterious woman in a plain white ancient-Chinese gown, with a delicate but unnaturally pale face, long jet-black hair worn loose with no ornaments, standing motionless with her hem reaching the floor.\n<Subject 2> is the man in <Picture 3>: an ancient Chinese constable in a dark ink-black long robe reaching his ankles, with a black headscarf tied at the back of his head, a resolute square face with strong brows, a sheathed saber at his left hip, a round wooden waist tag reading \"捕\" on his right hip, and black short boots.\n<Picture 1> is the scene reference for the dim old room and the composition anchor for [Shot 1]: five carved openwork screen panels standing in a row at the center, a round stone-top table carrying a single cold unlit candle with its bare wick exposed and no flame or glow anywhere on it — a wooden luohan bed on the right side, a paper wall scroll on the wall, and a heavy wooden door on the left.\n\nsummary:\n[reference generation] A panicked ancient constable (<Subject 2>) bursts through the door of a dim old Chinese room at night, braces the door shut, and speaks a single line of dialogue to the pale woman in white (<Subject 1>) who steps out from behind the five-panel carved screen; the only lit object in the room is one small oil lamp, the candle on the stone table stays unlit, no pursuer is ever shown or heard, and no music plays.\n\nretention_analysis:\n<Subject 1>: partially_preserved - the white gown, pale face and long black hair from <Picture 2> are retained; her walk from behind the screen, her head turn and her stillness in [Shot 4] and [Shot 5] are newly generated, and she is absent from [Shot 1] through [Shot 3].\n<Subject 2>: partially_preserved - the dark robe, black headscarf, saber and waist tag from <Picture 3> are retained; his panicked entrance, bracing of the door and spoken line are newly generated.\n<Picture 1>: partially_preserved - the room layout, the five-panel screen, the stone table with its cold unlit candle, the luohan bed and the wall scroll are retained as the setting; the lit oil lamp beside the luohan bed, its flame, the cast shadows and the night atmosphere are newly generated.\n\ndetailed_description:\nLive-action cinematic Chinese period horror in a dim old Chinese room at night. The room contains exactly one working light source: a small squat bronze oil lamp with a low steady open flame on the low cabinet beside the luohan bed at the mid-ground right; every other object is unlit, and all warm light comes only from that flame. [Shot 1] In the room shown in <Picture 1>, the heavy wooden door on the left wall is flung open and the constable (<Subject 2>) stumbles in, catching his balance while the camera tracks forward from the doorway behind him, his dark robe swallowing what little light reaches him. The five carved screen panels stand in the deep center of the frame, and the oil lamp's low flame is pressed flat for a moment by the draft he brings in. [Shot 2] At 00:01.200, the camera holds a static medium shot as the constable braces the door shut with both hands, his back to the room, the oil lamp flame ducking low again under the draft and slowly recovering. [Shot 3] At 00:02.000, the camera pulls back from the constable to reveal the whole room: the five openwork carved panels throw striped shadows onto the wall behind them, the stripes swaying with the lamp's flicker, and the round stone-top table comes into view with the candle on it completely dead — a white wax pillar in a bronze holder, bare wick exposed and cold, no flame, no halo of light around its tip, no pool of warm light on the tabletop — while the luohan bed on the right and the paper wall scroll also come into view, the constable a small foreground figure at the left. [Shot 4] At 00:03.200, the camera pushes in slowly toward the screen as the woman (<Subject 1>) steps out from behind the panels, her hem dragging and only her upper body moving, then stops in the lamplight and turns her head toward the constable in the left depth of the same frame; her pale face reads colder than the warm lamplight around it, no part of her is visible in any earlier shot, and the oil lamp stays inside the frame throughout, still the only lit object. [Shot 5] At 00:05.400, the camera trucks slowly rightward from the constable toward the woman: bracing the door without retreating, he lowers his voice and says <d>[Chinese] 对不起，我不是有意闯进来的，外面有东西在追我。</d> while the oil lamp flame trembles faintly with his voice; her white gown is the brightest shape in the room, and after the line both figures stand motionless facing each other across the screen panels, the shot ending on the two of them, the candle on the stone table still cold and dead. His saber stays sheathed for the whole clip, nothing and no one is ever shown or heard outside the closed door, and no second light source — no burning candle, no standing lantern — ever appears in the room.\n\noverall_soundscape:\nA heavy door slam and a muffled wooden groan open the clip under the constable's (<Subject 2>) rapid ragged panting, and a draft of night wind seeps in from outside the closed door. The single oil lamp contributes a faint wick crackle that continues under the whole clip. Soft cloth rustle and nearly silent footsteps mark the woman (<Subject 1>) stepping out from behind the screen. After the constable (<Subject 2>) finishes his line, the room falls into dead silence with only the faint flicker of the lamp flame.\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 448,
-    "height": 256,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788156878022,
-    "promptEn": "constable bursts in and blocks the door, talks to a pale woman behind the screen",
-    "lastPlayedAt": 1788327238271,
-    "playCount": 21
-  },
-  {
-    "id": "a0b752e7ac0d",
-    "source": "h3console",
-    "prompt": "捕快紧张害怕地说着话",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the single young male constable whose facial identity, black headscarf, dark indigo robe, leather waist belt, and period styling come from <Picture 1>. <Picture 1> is used only as an identity, clothing, hairstyle, and accessory reference; its multi-view character-sheet layout is not reproduced in the target video.\n\nsummary:\n[reference generation] The target video is a reusable silent speaking-performance clip for the finite expression state \"afraid\". It shows only <Subject 1> in a stable medium close-up, displaying restrained fear while making natural speech-related mouth movements that do not correspond to any specific sentence. The performance finishes close to its opening posture so the clip can repeat during variable-length TTS playback.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the same facial identity, black headscarf, dark indigo constable robe, leather belt, age, body proportions, hairstyle, and period styling from <Picture 1> remain consistent throughout the video.\n\ndetailed_description:\nThe target video uses a cinematic Chinese period-mystery style with realistic skin texture, restrained performance, soft warm key light, dim brown-black surroundings, and shallow depth of field. The output is a normal cinematic frame containing one character, never a character sheet, collage, split screen, turnaround view, or design-board layout.\n\n[Shot 1] A static medium close-up frames <Subject 1> from the upper chest upward. He stands nearly front-facing with his torso angled slightly toward an off-screen conversation partner. His facial identity, black headscarf, dark indigo robe collar, leather shoulder details, and period appearance follow <Picture 1>. The background is a softly blurred, dim old Chinese interior with no readable signs, no additional people, and no distinct story event. The camera remains completely stationary with no zoom, pan, tilt, tracking movement, reframing, focus shift, or cut.\n\nDuring the opening moment, <Subject 1> holds a guarded resting posture and looks toward the off-screen conversation partner. Restrained fear develops through slightly widened eyes, gently tightened eyebrows, shallow breathing, mild tension around the jaw, and a small swallow. The emotion remains believable and controlled, without theatrical grimacing, screaming, crying, or panic-driven body movement.\n\nHe begins natural speech-related mouth movement at a measured conversational pace. His lips open and close with varied, moderate articulation and brief pauses, but the movement does not match any identifiable sentence, language, or fixed phoneme sequence. His mouth never stretches into exaggerated shapes. While continuing the generic speaking motion, his eyes briefly glance toward the edge of the frame as if checking for a nearby threat, then return to the off-screen conversation partner. His head draws back by only a few centimeters and settles. His shoulders rise subtly with one quiet breath and lower again. His hands remain outside the frame, and he performs no semantic action that would bind the clip to a particular line of dialogue.\n\nIn the final portion, the mouth movement slows and closes naturally. The frightened alertness remains visible, while the gaze, head angle, shoulder height, facial tension, and framing return close to the opening state. The final image stays stable and visually compatible with replaying the clip from the beginning. There are no scene changes, extra characters, new objects, subtitles, logos, watermarks, visible text, or generated dialogue.\n\noverall_soundscape: N/A\n\nnon_diegetic_music: N/A",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788154942596,
-    "promptEn": "constable talking, looking scared",
-    "lastPlayedAt": 1788329531593,
-    "playCount": 21
-  },
-  {
-    "id": "b6608e2eac60",
-    "source": "h3console",
-    "prompt": "宇航员站在火星峡谷边，头盔反光照出外星废墟，镜头拉高俯瞰古城",
-    "fullPrompt": "Live-action, photorealistic, cinematic science fiction. Cold, low-saturation color grading with bluish shadows, and a thin dust haze softening every distant edge.\n\n[Shot 1] The camera pushes in slowly with small amplitude at slow speed toward an astronaut standing motionless at the rim of a Martian canyon, seen from behind and slightly to the side in a medium shot; his face is never shown at any point. His suit is matte off-white and light gray, with dark-gray corrugated bellows at the shoulders and knees, a rectangular dark-gray equipment module on the chest, and a rectangular life-support backpack with two vertical cylindrical tanks rising above his shoulders; the suit carries no flags, no national insignia, no badges and no legible lettering anywhere. His helmet is a one-piece spherical shell, and its full-arc visor carries a blue-gray mirrored coating that returns only reflections, never an interior view. Below his boots a ledge of wind-rippled sand and scattered dark bedrock slants toward the canyon floor, the opposite wall shows layered strata, and the horizon sits in the upper third of the frame. The sun sits low, fifteen to twenty degrees above the horizon behind the camera and out of frame, never a disc inside the shot; its pale, cool light lays long blue-shifted shadows across the sand, and no warm golden key light touches the suit or the sand. The sky is a cold dusty gray with a faint violet-gray cast, lightening to gray-blue overhead. The visor's mirror reflects only the silhouette of an abandoned alien city across the canyon: truncated pyramids, spiral shells, ribbed colonnades, and honeycomb clusters, with no rectangular curtain walls. As the push-in continues, the visor grows into a full-frame close-up and the reflected skyline becomes clearly legible; the reflection never contains a face.\n\n[Shot 2] At 00:06.000, the shot cuts to a high aerial view of that alien city. The camera descends slowly over the ruins, a pedestal down that loses altitude steadily instead of trucking sideways or pushing in. The buildings spread along the canyon wall, tens to hundreds of meters tall, larger at the top and breaking into smaller forms below, their surfaces wind-eroded, matte, in dull brown and gray-green. The canyon floor and its strata unfold layer by layer during the descent. The city is utterly inert: not one window or lamp is lit, no aircraft stands or flies, and nothing moves anywhere in the frame. The hour is still and windless; the only haze is one uniform layer that softens and desaturates the far edges, and no dust plume or column rises from any point on the canyon floor.\n\n[Shot 3] At 00:11.000, the camera holds a static shot on the empty city from a lower, closer position. The four simplified Chinese characters \"我们又来了\" appear at the center of the frame in thin white type. During the last second and a half the four characters fade out to nothing, fully gone while the city behind them is still faintly visible, and only then does the image finish fading to black.\n\noverall_soundscape: A sustained low-frequency Martian wind whistles across the canyon while steady breathing inside the helmet is audible during the first shot; from the aerial view the wind turns vast and distant, and the soundscape thins steadily toward silence. No voice of any kind is heard in the film.\n\nnon_diegetic_music: A low drone with sub-bass weight rises under the opening shot, thickens over the aerial descent, and recedes with the wind until it fades away with the picture.\n",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1788153988474,
-    "promptEn": "astronaut at a martian canyon, visor reflecting alien ruins, then aerial view of the city",
-    "lastPlayedAt": 1788329861863,
-    "playCount": 22
-  },
-  {
-    "id": "194fc2b56cb5",
-    "source": "h3console",
-    "prompt": "宇航员站在火星峡谷，头盔反光照出外星废城",
-    "fullPrompt": "Live-action, photorealistic, cinematic science fiction. Cold, low-saturation color grading with bluish shadows, and a thin dust haze softening every distant edge.\n\n[Shot 1] The camera pushes in slowly with small amplitude at slow speed toward an astronaut standing motionless at the rim of a Martian canyon, seen from behind and slightly to the side in a medium shot; his face is never shown at any point. His suit is matte off-white and light gray, with dark-gray corrugated bellows at the shoulders and knees, a rectangular dark-gray equipment module on the chest, and a rectangular life-support backpack with two vertical cylindrical tanks rising above his shoulders. His helmet is a one-piece spherical shell, and its full-arc visor carries a blue-gray mirrored coating that returns only reflections, never an interior view. Below his boots a ledge of wind-rippled sand and scattered dark bedrock slants toward the canyon floor, the opposite wall shows layered strata, and the horizon sits in the upper third of the frame. The sun hangs low, roughly fifteen to twenty degrees above the horizon, throwing long blue-shifted shadows across the sand, while the sky stays a muted rust-brown that fades to gray-blue overhead. The visor's mirror reflects only the silhouette of an abandoned alien city across the canyon: truncated pyramids, spiral shells, ribbed colonnades, and honeycomb clusters, with no rectangular curtain walls. As the push-in continues, the visor grows into a full-frame close-up and the reflected skyline becomes clearly legible; the reflection never contains a face.\n\n[Shot 2] At 00:06.000, the shot cuts to a high aerial view of that alien city. The camera descends slowly over the ruins, a pedestal down that loses altitude steadily instead of trucking sideways or pushing in. The buildings spread along the canyon wall, tens to hundreds of meters tall, larger at the top and breaking into smaller forms below, their surfaces wind-eroded, matte, in dull brown and gray-green. The canyon floor and its strata unfold layer by layer during the descent. The city is utterly inert: not one window or lamp is lit, no aircraft stands or flies, no smoke or dust column rises, and nothing moves anywhere in the frame, while distant structures lose their saturation into the haze.\n\n[Shot 3] At 00:11.000, the camera holds a static shot on the empty city from a lower, closer position. The four simplified Chinese characters \"我们又来了\" appear at the center of the frame in thin white type, and then the whole image slowly fades to black, the characters fading out with it.\n\noverall_soundscape: A sustained low-frequency Martian wind whistles across the canyon while steady breathing inside the helmet is audible during the first shot; from the aerial view the wind turns vast and distant, and the soundscape thins steadily toward silence. No voice of any kind is heard in the film.\n\nnon_diegetic_music: A low drone with sub-bass weight rises under the opening shot, thickens over the aerial descent, and recedes with the wind until it fades away with the picture.\n",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1788138492364,
-    "promptEn": "astronaut on a Martian canyon, helmet reflecting an alien ruined city",
-    "lastPlayedAt": 1788330744593,
-    "playCount": 22
-  },
-  {
-    "id": "f84b8f044ec5",
-    "source": "h3console",
-    "prompt": "捕快慌张闯进屋抵住门，跟屏风后的白衣女子说话",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the woman in <Picture 2>: a mysterious woman in a plain white ancient-Chinese gown, with a delicate but unnaturally pale face, long jet-black hair worn loose with no ornaments, standing motionless with her hem reaching the floor.\n<Subject 2> is the man in <Picture 3>: an ancient Chinese constable in a dark ink-black long robe reaching his ankles, with a black headscarf tied at the back of his head, a resolute square face with strong brows, a sheathed saber at his left hip, a round wooden waist tag reading \"捕\" on his right hip, and black short boots.\n<Picture 1> is the scene reference for the dim old room and the composition anchor for [Shot 1]: five carved openwork screen panels standing in a row at the center, a round stone-top table carrying a single cold unlit candle with its bare wick exposed and no flame or glow anywhere on it — a wooden luohan bed on the right side, a paper wall scroll on the wall, and a heavy wooden door on the left.\n\nsummary:\n[reference generation] A panicked ancient constable (<Subject 2>) bursts through the door of a dim old Chinese room at night, braces the door shut, and speaks a single line of dialogue to the pale woman in white (<Subject 1>) who steps out from behind the five-panel carved screen; the only lit object in the room is one small oil lamp, the candle on the stone table stays unlit, no pursuer is ever shown or heard, and no music plays.\n\nretention_analysis:\n<Subject 1>: partially_preserved - the white gown, pale face and long black hair from <Picture 2> are retained; her walk from behind the screen, her head turn and her stillness in [Shot 4] and [Shot 5] are newly generated, and she is absent from [Shot 1] through [Shot 3].\n<Subject 2>: partially_preserved - the dark robe, black headscarf, saber and waist tag from <Picture 3> are retained; his panicked entrance, bracing of the door and spoken line are newly generated.\n<Picture 1>: partially_preserved - the room layout, the five-panel screen, the stone table with its cold unlit candle, the luohan bed and the wall scroll are retained as the setting; the lit oil lamp beside the luohan bed, its flame, the cast shadows and the night atmosphere are newly generated.\n\ndetailed_description:\nLive-action cinematic Chinese period horror in a dim old Chinese room at night. The room contains exactly one working light source: a small squat bronze oil lamp with a low steady open flame on the low cabinet beside the luohan bed at the mid-ground right; every other object is unlit, and all warm light comes only from that flame. [Shot 1] In the room shown in <Picture 1>, the heavy wooden door on the left wall is flung open and the constable (<Subject 2>) stumbles in, catching his balance while the camera tracks forward from the doorway behind him, his dark robe swallowing what little light reaches him. The five carved screen panels stand in the deep center of the frame, and the oil lamp's low flame is pressed flat for a moment by the draft he brings in. [Shot 2] At 00:01.200, the camera holds a static medium shot as the constable braces the door shut with both hands, his back to the room, the oil lamp flame ducking low again under the draft and slowly recovering. [Shot 3] At 00:02.000, the camera pulls back from the constable to reveal the whole room: the five openwork carved panels throw striped shadows onto the wall behind them, the stripes swaying with the lamp's flicker, and the round stone-top table comes into view with the candle on it completely dead — a white wax pillar in a bronze holder, bare wick exposed and cold, no flame, no halo of light around its tip, no pool of warm light on the tabletop — while the luohan bed on the right and the paper wall scroll also come into view, the constable a small foreground figure at the left. [Shot 4] At 00:03.200, the camera pushes in slowly toward the screen as the woman (<Subject 1>) steps out from behind the panels, her hem dragging and only her upper body moving, then stops in the lamplight and turns her head toward the constable in the left depth of the same frame; her pale face reads colder than the warm lamplight around it, no part of her is visible in any earlier shot, and the oil lamp stays inside the frame throughout, still the only lit object. [Shot 5] At 00:05.400, the camera trucks slowly rightward from the constable toward the woman: bracing the door without retreating, he lowers his voice and says <d>[Chinese] 对不起，我不是有意闯进来的，外面有东西在追我。</d> while the oil lamp flame trembles faintly with his voice; her white gown is the brightest shape in the room, and after the line both figures stand motionless facing each other across the screen panels, the shot ending on the two of them, the candle on the stone table still cold and dead. His saber stays sheathed for the whole clip, nothing and no one is ever shown or heard outside the closed door, and no second light source — no burning candle, no standing lantern — ever appears in the room.\n\noverall_soundscape:\nA heavy door slam and a muffled wooden groan open the clip under the constable's (<Subject 2>) rapid ragged panting, and a draft of night wind seeps in from outside the closed door. The single oil lamp contributes a faint wick crackle that continues under the whole clip. Soft cloth rustle and nearly silent footsteps mark the woman (<Subject 1>) stepping out from behind the screen. After the constable (<Subject 2>) finishes his line, the room falls into dead silence with only the faint flicker of the lamp flame.\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 10,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788120171725,
-    "promptEn": "constable bursts in and blocks the door, talks to a pale woman behind the screen",
-    "lastPlayedAt": 1788329691719,
-    "playCount": 21
-  },
-  {
-    "id": "5c364152e54c",
-    "source": "h3console",
-    "prompt": "捕快慌张闯进屋抵住门，跟屏风后的白衣女子说话",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the woman in <Picture 2>: a mysterious woman in a plain white ancient-Chinese gown, with a delicate but unnaturally pale face, long jet-black hair worn loose with no ornaments, standing motionless with her hem reaching the floor.\n<Subject 2> is the man in <Picture 3>: an ancient Chinese constable in a dark ink-black long robe reaching his ankles, with a black headscarf tied at the back of his head, a resolute square face with strong brows, a sheathed saber at his left hip, a round wooden waist tag reading \"捕\" on his right hip, and black short boots.\n<Picture 1> is the scene reference for the dim old room and the composition anchor for [Shot 1]: five carved openwork screen panels standing in a row at the center, a round stone-top table carrying a single cold unlit candle with its bare wick exposed and no flame or glow anywhere on it — a wooden luohan bed on the right side, a paper wall scroll on the wall, and a heavy wooden door on the left.\n\nsummary:\n[reference generation] A panicked ancient constable (<Subject 2>) bursts through the door of a dim old Chinese room at night, braces the door shut, and speaks a single line of dialogue to the pale woman in white (<Subject 1>) who steps out from behind the five-panel carved screen; the only lit object in the room is one small oil lamp, the candle on the stone table stays unlit, no pursuer is ever shown or heard, and no music plays.\n\nretention_analysis:\n<Subject 1>: partially_preserved - the white gown, pale face and long black hair from <Picture 2> are retained; her walk from behind the screen, her head turn and her stillness in [Shot 4] and [Shot 5] are newly generated, and she is absent from [Shot 1] through [Shot 3].\n<Subject 2>: partially_preserved - the dark robe, black headscarf, saber and waist tag from <Picture 3> are retained; his panicked entrance, bracing of the door and spoken line are newly generated.\n<Picture 1>: partially_preserved - the room layout, the five-panel screen, the stone table with its cold unlit candle, the luohan bed and the wall scroll are retained as the setting; the lit oil lamp beside the luohan bed, its flame, the cast shadows and the night atmosphere are newly generated.\n\ndetailed_description:\nLive-action cinematic Chinese period horror in a dim old Chinese room at night. The room contains exactly one working light source: a small squat bronze oil lamp with a low steady open flame on the low cabinet beside the luohan bed at the mid-ground right; every other object is unlit, and all warm light comes only from that flame. [Shot 1] In the room shown in <Picture 1>, the heavy wooden door on the left wall is flung open and the constable (<Subject 2>) stumbles in, catching his balance while the camera tracks forward from the doorway behind him, his dark robe swallowing what little light reaches him. The five carved screen panels stand in the deep center of the frame, and the oil lamp's low flame is pressed flat for a moment by the draft he brings in. [Shot 2] At 00:01.200, the camera holds a static medium shot as the constable braces the door shut with both hands, his back to the room, the oil lamp flame ducking low again under the draft and slowly recovering. [Shot 3] At 00:02.000, the camera pulls back from the constable to reveal the whole room: the five openwork carved panels throw striped shadows onto the wall behind them, the stripes swaying with the lamp's flicker, and the round stone-top table comes into view with the candle on it completely dead — a white wax pillar in a bronze holder, bare wick exposed and cold, no flame, no halo of light around its tip, no pool of warm light on the tabletop — while the luohan bed on the right and the paper wall scroll also come into view, the constable a small foreground figure at the left. [Shot 4] At 00:03.200, the camera pushes in slowly toward the screen as the woman (<Subject 1>) steps out from behind the panels, her hem dragging and only her upper body moving, then stops in the lamplight and turns her head toward the constable in the left depth of the same frame; her pale face reads colder than the warm lamplight around it, no part of her is visible in any earlier shot, and the oil lamp stays inside the frame throughout, still the only lit object. [Shot 5] At 00:05.400, the camera trucks slowly rightward from the constable toward the woman: bracing the door without retreating, he lowers his voice and says <d>[Chinese] 对不起，我不是有意闯进来的，外面有东西在追我。</d> while the oil lamp flame trembles faintly with his voice; her white gown is the brightest shape in the room, and after the line both figures stand motionless facing each other across the screen panels, the shot ending on the two of them, the candle on the stone table still cold and dead. His saber stays sheathed for the whole clip, nothing and no one is ever shown or heard outside the closed door, and no second light source — no burning candle, no standing lantern — ever appears in the room.\n\noverall_soundscape:\nA heavy door slam and a muffled wooden groan open the clip under the constable's (<Subject 2>) rapid ragged panting, and a draft of night wind seeps in from outside the closed door. The single oil lamp contributes a faint wick crackle that continues under the whole clip. Soft cloth rustle and nearly silent footsteps mark the woman (<Subject 1>) stepping out from behind the screen. After the constable (<Subject 2>) finishes his line, the room falls into dead silence with only the faint flicker of the lamp flame.\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 10,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788120171608,
-    "promptEn": "constable bursts in and blocks the door, talks to a pale woman behind the screen",
-    "lastPlayedAt": 1788327580527,
-    "playCount": 21
-  },
-  {
-    "id": "b8c762a3672e",
-    "source": "h3console",
-    "prompt": "捕快慌张闯进屋抵住门，跟屏风后的白衣女子说话",
-    "fullPrompt": "subject_definitions:\n<Subject 1> is the woman in <Picture 2>: a mysterious woman in a plain white ancient-Chinese gown, with a delicate but unnaturally pale face, long jet-black hair worn loose with no ornaments, standing motionless with her hem reaching the floor.\n<Subject 2> is the man in <Picture 3>: an ancient Chinese constable in a dark ink-black long robe reaching his ankles, with a black headscarf tied at the back of his head, a resolute square face with strong brows, a sheathed saber at his left hip, a round wooden waist tag reading \"捕\" on his right hip, and black short boots.\n<Picture 1> is the scene reference for the dim old room and the composition anchor for [Shot 1]: five carved openwork screen panels standing in a row at the center, a round stone-top table with one unlit candle, a wooden luohan bed on the right side, a paper wall scroll on the wall, and a heavy wooden door on the left.\n\nsummary:\n[reference generation] A panicked ancient constable (<Subject 2>) bursts through the door of a dim lamplit old Chinese room at night, braces the door shut, and speaks a single line of dialogue to the pale woman in white (<Subject 1>) who steps out from behind the five-panel carved screen; no pursuer is ever shown or heard, and no music plays.\n\nretention_analysis:\n<Subject 1>: partially_preserved - the white gown, pale face and long black hair from <Picture 2> are retained; her walk from behind the screen, her head turn and her stillness in [Shot 4] and [Shot 5] are newly generated, and she is absent from [Shot 1] through [Shot 3].\n<Subject 2>: partially_preserved - the dark robe, black headscarf, saber and waist tag from <Picture 3> are retained; his panicked entrance, bracing of the door and spoken line are newly generated.\n<Picture 1>: partially_preserved - the room layout, the five-panel screen, the stone table with unlit candle, the luohan bed and the wall scroll are retained as the setting; the moving flame, the cast shadows and the night atmosphere are newly generated.\n\ndetailed_description:\nLive-action cinematic Chinese period horror, lit entirely by a single warm oil lamp inside a dim old room. [Shot 1] In the room shown in <Picture 1>, the heavy wooden door on the left wall is flung open from outside and the constable (<Subject 2>) stumbles in, catching his balance after two staggering steps while the camera tracks forward from the doorway behind him, his dark robe swallowing what little light reaches him. The five carved screen panels stand in the deep center of the frame, and the small oil lamp burns at the mid-ground right; the flame is pressed low for a moment by the draft he brings in. [Shot 2] At 00:01.200, the camera holds a static medium shot as the constable braces the door shut with both hands behind him, his back turned to the room, the lamp flame ducking low again under the draft and slowly recovering. [Shot 3] At 00:02.000, the camera pulls back from the constable to reveal the whole room: the five openwork carved panels throw striped shadows onto the wall behind them, those stripes sway almost imperceptibly with the flicker, and the round stone-top table with its unlit candle, the wooden luohan bed on the right, and the paper wall scroll come into view while he stays a small foreground figure at the left, his breathing slowing. [Shot 4] At 00:03.200, the camera pushes in slowly toward the screen as the woman (<Subject 1>) steps out from behind the panels, her hem dragging and only her upper body moving, then stops in the lamplight and turns her head toward the constable, who remains in the left depth of the same frame; her pale face reads colder than the warm lamplight around it, no part of her is visible in any earlier shot, and the oil lamp stays inside the frame throughout. [Shot 5] At 00:05.400, the camera trucks slowly rightward from the constable toward the woman: bracing the door without retreating, he lowers his voice and says <d>[Chinese] 对不起，我不是有意闯进来的，外面有东西在追我。</d> while the lamp flame trembles faintly with his voice; her white gown is the brightest shape in the room, and after the line both figures stand motionless, facing each other across the screen panels, and the shot ends holding on the two of them. His saber stays sheathed for the whole clip, nothing and no one is ever shown or heard outside the closed door, and no second light source ever appears in the room.\n\noverall_soundscape:\nA heavy door slam and a muffled wooden groan open the clip under the constable's (<Subject 2>) rapid ragged panting, and a draft of night wind seeps in from outside the closed door. The single oil lamp contributes a faint wick crackle that continues under the whole clip. Soft cloth rustle and nearly silent footsteps mark the woman (<Subject 1>) stepping out from behind the screen. After the constable (<Subject 2>) finishes his line, the room falls into dead silence with only the faint flicker of the lamp flame.\n\nnon_diegetic_music:\nN/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 10,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788106707982,
-    "promptEn": "constable bursts in and blocks the door, talks to a pale woman behind the screen",
-    "lastPlayedAt": 1788330860686,
-    "playCount": 22
-  },
-  {
-    "id": "1fa40bca47f5",
-    "source": "h3console",
-    "prompt": "捕快站定不动，小声说自己不会出声，屏风后女子静立不动",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (https://shadiao-2d.oss-cn-beijing.aliyuncs.com/drama3/visual-game/packs/painted_skin_room/assets/room_reference/90f0634b250adb10fd88292af864d590ed444eef458a67dd8a084492e57e2429.png) (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman shown in <Picture 2> (https://shadiao-drama.oss-cn-beijing.aliyuncs.com/visual-game/packs/painted_skin_room/assets/model_sheet_face/0e1bff54b045db70b3f73b93e0b1187759568a48d8e8bc510638cfc7dd6eb20a.png), with long black hair falling past her shoulders, a flowing white traditional Chinese robe, and a delicately featured face with an unnaturally pale complexion.\n<Subject 2> is the young male constable shown in <Picture 3> (https://shadiao-drama.oss-cn-beijing.aliyuncs.com/visual-game/packs/painted_skin_room/assets/constable_model_sheet_face/7c8c0edc291bc7db1d5fa4c12fce867f0b0a8574863a02de295e84456b0f9f9d.png), with a strong, resolute face, wearing a dark indigo robe, a small round badge reading \"捕\" at his waist, a dao sword at his hip, black boots, and a black head cloth.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens standing across the center, a round stone table bearing a single unlit candle, a luohan couch along the right wall, a hanging scroll on the wall, and deep shadow pools across the plaster walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> holding his ground in the dim room established by <Picture 1> and quietly stating that he will stay where he is and make no sound, then keeping completely still and silent, while <Subject 1> remains motionless behind the carved wooden screens and the oil lamp flame keeps its low, steady sway.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved screens, round stone table with the unlit candle, luohan couch, hanging scroll, and amber lamp shadows are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the white robe, long black hair, pale complexion, and still posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark indigo robe, the waist badge, the dao sword, the black boots, and the head cloth are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with a single warm amber light source set against deep shadow and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room: five carved wooden screens standing across the center, a round stone table holding a single unlit candle, a luohan couch along the right wall, and a hanging scroll flat against the plaster wall. The only light in the room comes from a small oil lamp on a low table, its flame swaying gently and throwing slow-moving amber pools across the stone floor, while the candle on the round table stands unlit. <Subject 2>, the young constable in a dark indigo robe with a round badge reading \"捕\" at his waist, a dao sword at his hip, black boots, and a black head cloth, stands planted in the foreground with his weight evenly on both feet, facing the screens. Through the narrow vertical gaps between the carved panels, <Subject 1>, the woman in a flowing white robe with long black hair and an unnaturally pale, delicate face, is partially visible and perfectly motionless. The camera holds a static shot. <Subject 2> lowers his chin slightly and speaks in a low, level voice, [Chinese] 我就在这里待着，不会出声。 His right hand stays open at his side, away from the dao hilt, and his feet do not shift.\n[Shot 2] At 00:04.500, the camera cuts to a medium close-up of <Subject 2> from a slightly low angle. The amber lamp light rises along one side of his jaw and leaves the rest of his face in shadow. His lips are closed as the last syllable settles, and his breathing slows to a barely visible rise of his chest. His eyes stay fixed on the narrow gap between the two central screens where the pale shape of <Subject 1> stands. The camera pushes in with small amplitude at slow speed. The lamp flame bends once, deepening the shadow along the carved lattice, and the room settles into complete stillness with neither figure moving.\n\noverall_soundscape:\nA faint continuous sputter from the oil lamp wick sits under a low room tone. Silk shifts once behind the screens, followed by a single soft creak of old wood. The constable's breathing settles into near silence.\n\nnon_diegetic_music:\nA single sustained low string drone at a very slow tempo, holding without swell beneath the scene.",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1788085282201,
-    "promptEn": "constable stands still, quietly says he won't make a sound, woman stays silent behind the screen",
-    "lastPlayedAt": 1788330428360,
-    "playCount": 22
-  },
-  {
-    "id": "683ab2fd4681",
-    "source": "h3console",
-    "prompt": "捕快站定不动，小声说自己不会出声，屏风后女子静立不动",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman shown in <Picture 2>, with long black hair falling past her shoulders, a flowing white traditional Chinese robe, and a delicately featured face with an unnaturally pale complexion.\n<Subject 2> is the young male constable shown in <Picture 3>, with a strong, resolute face, wearing a dark indigo robe, a small round badge reading \"捕\" at his waist, a dao sword at his hip, black boots, and a black head cloth.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens standing across the center, a round stone table bearing a single unlit candle, a luohan couch along the right wall, a hanging scroll on the wall, and deep shadow pools across the plaster walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> holding his ground in the dim room established by <Picture 1> and quietly stating that he will stay where he is and make no sound, then keeping completely still and silent, while <Subject 1> remains motionless behind the carved wooden screens and the oil lamp flame keeps its low, steady sway.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved screens, round stone table with the unlit candle, luohan couch, hanging scroll, and amber lamp shadows are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the white robe, long black hair, pale complexion, and still posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark indigo robe, the waist badge, the dao sword, the black boots, and the head cloth are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with a single warm amber light source set against deep shadow and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room: five carved wooden screens standing across the center, a round stone table holding a single unlit candle, a luohan couch along the right wall, and a hanging scroll flat against the plaster wall. The only light in the room comes from a small oil lamp on a low table, its flame swaying gently and throwing slow-moving amber pools across the stone floor, while the candle on the round table stands unlit. <Subject 2>, the young constable in a dark indigo robe with a round badge reading \"捕\" at his waist, a dao sword at his hip, black boots, and a black head cloth, stands planted in the foreground with his weight evenly on both feet, facing the screens. Through the narrow vertical gaps between the carved panels, <Subject 1>, the woman in a flowing white robe with long black hair and an unnaturally pale, delicate face, is partially visible and perfectly motionless. The camera holds a static shot. <Subject 2> lowers his chin slightly and speaks in a low, level voice, [Chinese] 我就在这里待着，不会出声。 His right hand stays open at his side, away from the dao hilt, and his feet do not shift.\n[Shot 2] At 00:04.500, the camera cuts to a medium close-up of <Subject 2> from a slightly low angle. The amber lamp light rises along one side of his jaw and leaves the rest of his face in shadow. His lips are closed as the last syllable settles, and his breathing slows to a barely visible rise of his chest. His eyes stay fixed on the narrow gap between the two central screens where the pale shape of <Subject 1> stands. The camera pushes in with small amplitude at slow speed. The lamp flame bends once, deepening the shadow along the carved lattice, and the room settles into complete stillness with neither figure moving.\n\noverall_soundscape:\nA faint continuous sputter from the oil lamp wick sits under a low room tone. Silk shifts once behind the screens, followed by a single soft creak of old wood. The constable's breathing settles into near silence.\n\nnon_diegetic_music:\nA single sustained low string drone at a very slow tempo, holding without swell beneath the scene.",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788071067702,
-    "promptEn": "constable stands still, quietly says he won't make a sound, woman stays silent behind the screen",
-    "lastPlayedAt": 1788329657693,
-    "playCount": 21
-  },
-  {
-    "id": "2e09fcd3d42c",
-    "source": "h3console",
-    "prompt": "捕快站定不动，小声说自己不会出声，屏风后女子静立不动",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman shown in <Picture 2>, with long black hair falling past her shoulders, a flowing white traditional Chinese robe, and a delicately featured face with an unnaturally pale complexion.\n<Subject 2> is the young male constable shown in <Picture 3>, with a strong, resolute face, wearing a dark indigo robe, a small round badge reading \"捕\" at his waist, a dao sword at his hip, black boots, and a black head cloth.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens standing across the center, a round stone table bearing a single unlit candle, a luohan couch along the right wall, a hanging scroll on the wall, and deep shadow pools across the plaster walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> holding his ground in the dim room established by <Picture 1> and quietly stating that he will stay where he is and make no sound, then keeping completely still and silent, while <Subject 1> remains motionless behind the carved wooden screens and the oil lamp flame keeps its low, steady sway.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved screens, round stone table with the unlit candle, luohan couch, hanging scroll, and amber lamp shadows are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the white robe, long black hair, pale complexion, and still posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark indigo robe, the waist badge, the dao sword, the black boots, and the head cloth are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with a single warm amber light source set against deep shadow and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room: five carved wooden screens standing across the center, a round stone table holding a single unlit candle, a luohan couch along the right wall, and a hanging scroll flat against the plaster wall. The only light in the room comes from a small oil lamp on a low table, its flame swaying gently and throwing slow-moving amber pools across the stone floor, while the candle on the round table stands unlit. <Subject 2>, the young constable in a dark indigo robe with a round badge reading \"捕\" at his waist, a dao sword at his hip, black boots, and a black head cloth, stands planted in the foreground with his weight evenly on both feet, facing the screens. Through the narrow vertical gaps between the carved panels, <Subject 1>, the woman in a flowing white robe with long black hair and an unnaturally pale, delicate face, is partially visible and perfectly motionless. The camera holds a static shot. <Subject 2> lowers his chin slightly and speaks in a low, level voice, <d>[Chinese] 我就在这里待着，不会出声。</d> His right hand stays open at his side, away from the dao hilt, and his feet do not shift.\n[Shot 2] At 00:04.500, the camera cuts to a medium close-up of <Subject 2> from a slightly low angle. The amber lamp light rises along one side of his jaw and leaves the rest of his face in shadow. His lips are closed as the last syllable settles, and his breathing slows to a barely visible rise of his chest. His eyes stay fixed on the narrow gap between the two central screens where the pale shape of <Subject 1> stands. The camera pushes in with small amplitude at slow speed. The lamp flame bends once, deepening the shadow along the carved lattice, and the room settles into complete stillness with neither figure moving.\n\noverall_soundscape:\nA faint continuous sputter from the oil lamp wick sits under a low room tone. Silk shifts once behind the screens, followed by a single soft creak of old wood. The constable's breathing settles into near silence.\n\nnon_diegetic_music:\nA single sustained low string drone at a very slow tempo, holding without swell beneath the scene.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788062322252,
-    "promptEn": "constable stands still, quietly says he won't make a sound, woman stays silent behind the screen",
-    "lastPlayedAt": 1788330674528,
-    "playCount": 22
-  },
-  {
-    "id": "12ea9ce00f36",
-    "source": "h3console",
-    "prompt": "银发偶像在后台化妆间，红灯闪烁，递出血红契约",
-    "fullPrompt": "Image-to-video opening, 4 seconds. Use the uploaded first-frame reference as the only identity source. Preserve the Korean-webtoon illustrated silver-white long-haired androgynous male idol, same stage costume, same debut-night backstage dressing room and vanity mirror. Begin from that frame with the NPC large and clearly identifiable; only add red stage-light pulse and a slow push-in as he offers the blood-red contract. No subtitles, no readable text, no extra people, no UI, no watermark.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 4,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788039783070,
-    "promptEn": "silver-haired idol backstage, red light pulsing, offering a blood-red contract",
-    "lastPlayedAt": 1788330568442,
-    "playCount": 22
-  },
-  {
-    "id": "75bdbacd3012",
-    "source": "h3console",
-    "prompt": "银发偶像在后台化妆间，红灯闪烁，递出血红契约",
-    "fullPrompt": "Image-to-video opening, 4 seconds. Use the uploaded first-frame reference as the only identity source. Preserve the Korean-webtoon illustrated silver-white long-haired androgynous male idol, same stage costume, same debut-night backstage dressing room and vanity mirror. Begin from that frame with the NPC large and clearly identifiable; only add red stage-light pulse and a slow push-in as he offers the blood-red contract. No subtitles, no readable text, no extra people, no UI, no watermark.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 4,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788039782992,
-    "promptEn": "silver-haired idol backstage, red light pulsing, offering a blood-red contract",
-    "lastPlayedAt": 1788329627669,
-    "playCount": 21
-  },
-  {
-    "id": "69e75a6b3997",
-    "source": "h3console",
-    "prompt": "受伤猎人站在废弃地铁副本入口，闪电划过，指向隧道",
-    "fullPrompt": "Image-to-video opening, 4 seconds. Use the uploaded first-frame reference as the exact identity/style source. Preserve the Korean-webtoon illustrated wounded S-rank hunter, same tactical costume, same ruined subway dungeon entrance and broken turnstile. Begin from that frame with the NPC large and clearly identifiable; only add lightning flicker, breath, and a small command gesture toward the tunnel. No subtitles, no readable text, no extra people, no UI, no watermark.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 4,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788027833836,
-    "promptEn": "wounded hunter at a ruined subway dungeon entrance, lightning flickers, pointing into the tunnel",
-    "lastPlayedAt": 1788330814648,
-    "playCount": 22
-  },
-  {
-    "id": "cde399c27dc4",
-    "source": "h3console",
-    "prompt": "神明戴金冠披红盖头，月光婚房里静静打量你",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> from [Shot 1] is fully referenced.\n\nintegrated_multimodal_description: [Shot 1] Oriental fantasy visual-novel illustration, image-to-video continuation from <Picture 1>. Preserve the exact deity identity from <Picture 1>: tall ethereal active NPC, very long white hair, pale divine face, ornate gold crown and golden ornaments, layered white ceremonial robe seen through translucent red bridal veil. Preserve the fixed moonlit temple bridal chamber and ritual scene: red wedding veils, altar table with ritual objects, candles and lanterns, open moonlit garden beyond, jade contract and red bridal cover in the foreground. Begin with the deity clearly visible in the first beat behind the red veil, calmly judging the viewer. The camera performs a slow small-amplitude Push In; red gauze sways gently, candle flames tremble, moonlight glints on the crown and jade contract, and the deity lowers the gaze slightly without changing pose or costume. Do not add another character, do not change face, hair, clothing, scene layout, or painterly style. No subtitles, no readable text, no UI, no watermark, no logo.\noverall_soundscape: Soft veil rustle, tiny candle crackle, distant night wind through the temple doors, and a faint ritual bell resonance.\nnon_diegetic_music: Slow sacred ambient strings and low choral pad, wordless and restrained.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 4,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788026877547,
-    "promptEn": "a deity in gold crown and red veil, quietly judging you in a moonlit bridal chamber",
-    "lastPlayedAt": 1788330126093,
-    "playCount": 22
-  },
-  {
-    "id": "d5f811737e95",
-    "source": "h3console",
-    "prompt": "公寓管理员站在诡异走廊，红色电梯灯闪，递出钥匙",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> from [Shot 1] is fully referenced.\n\nintegrated_multimodal_description: [Shot 1] Korean webtoon visual-novel illustration, image-to-video continuation from <Picture 1>. Preserve the exact apartment manager identity from <Picture 1>: young adult East Asian man, neat black side-parted hair, pale calm face, dark manager jacket over white shirt and black tie, reserved dangerous smile. Preserve the fixed strange apartment corridor: dark midnight hallway, elevator on the right glowing red, long wet reflective floor, ceiling fluorescent lights, front desk edge and key in the foreground. Begin with the active NPC clearly visible in the first beat, standing at the desk and extending the key toward the viewer. The camera performs a slow small-amplitude Push In while the red elevator indicator flickers once, the key ring makes a tiny swing, his jacket hem and hair move subtly, and cold corridor light breathes across the wall. Do not add another character, do not change gender, clothing, face, hairstyle, scene layout, or art style. No subtitles, no readable text, no UI, no watermark, no logo.\noverall_soundscape: Soft fluorescent hum, distant elevator machinery, faint key metal clink, and quiet apartment corridor air.\nnon_diegetic_music: Low restrained suspense drone, very soft and without lyrics.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 4,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788026876886,
-    "promptEn": "apartment manager in an eerie corridor, red elevator light flickering, holding out a key",
-    "lastPlayedAt": 1788331044844,
-    "playCount": 22
-  },
-  {
-    "id": "cdd03cca8461",
-    "source": "h3console",
-    "prompt": "缝脸长发怨灵站在地铁站台，灯光闪烁，露出诡异笑容",
-    "fullPrompt": "Use the reference image as exact first frame and identity anchor. The stitched-face long-haired curse in dark clothes is the only visual subject, face and upper body clear, same cold subway platform. Slow push-in, fluorescent lights flicker, papers move, he tilts his head with a cruel playful smile. Background crowd static only. No subtitles, no readable text, no UI, no logo, no watermark, no title card, no black frame.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 4,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788026569422,
-    "promptEn": "stitched-face long-haired curse on a subway platform, flickering lights, cruel smile",
-    "lastPlayedAt": 1788330238191,
-    "playCount": 22
-  },
-  {
-    "id": "b8dd0f62f648",
-    "source": "h3console",
-    "prompt": "苍白梦魔穿着正装，在梦境列车车厢里伸出手",
-    "fullPrompt": "Use the reference image as exact first frame and identity anchor. The pale smiling dream demon in dark formal clothes is the only visual subject, face and upper body clear, same blue-purple train carriage. Slow push-in, dream mist curls, he slowly extends one hand. Sleeping passengers are static background only. No subtitles, no readable text, no UI, no logo, no watermark, no title card, no black frame.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 4,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788026569123,
-    "promptEn": "pale dream demon in a suit, reaching out a hand inside a dream train car",
-    "lastPlayedAt": 1788330294235,
-    "playCount": 22
-  },
-  {
-    "id": "bd6e2b264243",
-    "source": "h3console",
-    "prompt": "金发士兵全副武装，在废弃地下室战场守着入口",
-    "fullPrompt": "Use the reference image as exact first frame and identity anchor. The blond soldier in combat gear is the only visual subject, face or upper body clear and identifiable, same broken basement battlefield. Slow push-in, smoke moves, he braces in front of the entrance with tense warrior posture. No new performer, distant figures static only. No subtitles, no readable text, no UI, no logo, no watermark, no title card, no black frame.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 4,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788026568815,
-    "promptEn": "blond soldier in combat gear, bracing at the entrance of a ruined basement battlefield",
-    "lastPlayedAt": 1788330168134,
-    "playCount": 22
-  },
-  {
-    "id": "b0d6af33eace",
-    "source": "h3console",
-    "prompt": "橙发反派披着红云黑袍，在废村里抬手",
-    "fullPrompt": "Use the reference image as exact first frame and identity anchor. The orange-haired robed antagonist with facial piercings and black cloak with red cloud pattern is the only visual subject, face and upper body clear, same ruined village scene. Slow push-in, dust drifts, he raises one hand calmly. No new performer, background figures static only. No subtitles, no readable text, no UI, no logo, no watermark, no title card, no black frame.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 4,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788026568516,
-    "promptEn": "orange-haired villain in a black cloak with red clouds, raising a hand in a ruined village",
-    "lastPlayedAt": 1788327564525,
-    "playCount": 21
-  },
-  {
-    "id": "b077fd090415",
-    "source": "h3console",
-    "prompt": "白袍熔岩指挥官在战场上举起岩浆拳头",
-    "fullPrompt": "Use the reference image as exact first frame and identity anchor. The magma military commander in a white admiral coat is the only visual subject, face and upper body clear, same anime style and same battlefield. Slow push-in, smoke and magma glow move subtly, he raises one magma fist. No new performer, background crowd static only. No subtitles, no readable text, no UI, no logo, no watermark, no title card, no black frame.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 4,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788026568242,
-    "promptEn": "magma commander in a white coat, raising a lava fist on the battlefield",
-    "lastPlayedAt": 1788329191322,
-    "playCount": 21
-  },
-  {
-    "id": "1442013e94f0",
-    "source": "h3console",
-    "prompt": "黑发红袍王子在审判厅按下处决令",
-    "fullPrompt": "Image-to-video opening, 4 seconds. Use the uploaded first-frame reference as the exact identity/style source. Preserve the Korean-webtoon illustrated black-haired red-cloak prince, same royal trial hall, same execution-platform lighting and costume. Begin from that frame with the NPC large and clearly identifiable; only add candle flicker, cloak movement, and a slow push-in while he presses the execution order. No subtitles, no readable text, no extra people, no UI, no watermark.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 4,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788026068497,
-    "promptEn": "black-haired prince in a red cloak, pressing down an execution order in the trial hall",
-    "lastPlayedAt": 1788327356363,
-    "playCount": 21
-  },
-  {
-    "id": "0a5e37bb328c",
-    "source": "h3console",
-    "prompt": "Vecna在颠倒世界巢穴里，伸出爪子逼近",
-    "fullPrompt": "4-second H3 image-referenced video. Use the reference image for the first visible frame, identity, body texture, and pose continuity. Show Vecna as the sole dominant active NPC from frame one: tall humanoid monster with exposed vine-like red flesh and skull-like face, upper body clearly visible, reaching forward with one clawed hand. Fixed scene: Vecna psychic lair in the Upside Down, red lightning, twisted roots, smoky memory fragments, dark red haze. Action: Vecna leans forward and tightens his raised hand as if invading the viewer mind. No subtitles, no UI, no watermark, no logos, no readable text.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 4,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788025642732,
-    "promptEn": "Vecna reaching out a claw in his lair in the Upside Down, red lightning",
-    "lastPlayedAt": 1788328340709,
-    "playCount": 21
-  },
-  {
-    "id": "00862f966c3f",
-    "source": "h3console",
-    "prompt": "伏地魔在墓地举着魔杖，一步步逼近",
-    "fullPrompt": "4-second H3 image-referenced video. Use the reference image for the first visible frame, identity, costume, and pose continuity. Show Voldemort as the sole dominant active NPC from frame one: pale bald dark wizard in a black robe, snake-like face, wand raised, upper body clearly visible in the Little Hangleton graveyard after a dark resurrection ritual. Fixed scene: gothic tombstones, ritual cauldron glow, fog, blue-green moonlit darkness, broken graveyard ground. Action: Voldemort steps closer and lowers the wand toward the viewer in a cold threat. No subtitles, no UI, no watermark, no logos, no readable text.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 4,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788025640972,
-    "promptEn": "Voldemort in a graveyard, wand raised, stepping closer",
-    "lastPlayedAt": 1788328306678,
-    "playCount": 21
-  },
-  {
-    "id": "c3875fbd6cc4",
-    "source": "h3console",
-    "prompt": "无限旅馆走廊尽头的血盆大口墙，缓缓张大",
-    "fullPrompt": "4-second image-to-video. Use the uploaded first frame as the exact opening composition: the current active NPC Eternity Devil is the grotesque living wall mouth at the end of an infinite hotel corridor, many fleshy faces embedded in the walls, huge teeth and red maw clearly visible. Fixed corridor scene must remain consistent: repeating doors, red lamps, dark ceiling lights, no humans in foreground. Motion: slow dolly forward, corridor lamps flicker, wall flesh pulses slightly, the mouth opens a little wider. One monster subject only; no subtitles, no readable text, no UI, no watermark, no logos, no black frame.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 4,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788025632028,
-    "promptEn": "a giant mouth embedded in the wall at the end of an endless hotel corridor, slowly opening",
-    "lastPlayedAt": 1788330852679,
-    "playCount": 22
-  },
-  {
-    "id": "1c6e2df243f4",
-    "source": "h3console",
-    "prompt": "白发反派站在废墟街头，咧嘴笑着伸手",
-    "fullPrompt": "4-second image-to-video. Use the uploaded first frame as the exact opening composition: the current active NPC Tomura-like white-haired villain is large and clear in the foreground of a ruined night street, long pale hair, cracked smile, red torn cloak, bare scarred chest, one hand extended while dust and decay break the ground. Fixed rubble street scene and costume must remain consistent. Motion: slow camera push-in, dust swirls, cloak and hair flutter, the extended hand flexes. One active NPC only; no background co-performer; no subtitles, no readable text, no UI, no watermark, no logos, no black frame.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 4,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788025630249,
-    "promptEn": "white-haired villain in a ruined street at night, cracked smile, reaching out a hand",
-    "lastPlayedAt": 1788330856684,
-    "playCount": 22
-  },
-  {
-    "id": "d735f1fc010f",
-    "source": "h3console",
-    "prompt": "猫耳白发守卫蹲在废弃宫殿里，周身紫色气场",
-    "fullPrompt": "4-second image-to-video. Use the uploaded first frame as the exact opening composition: the current active NPC Neferpitou-like cat humanoid guard is clear in the foreground, white curly hair, cat ears, magenta eyes, dark outfit, crouched low with claws on the palace floor, violet aura behind. Fixed ruined palace room at night must remain consistent. Motion: subtle low camera push-in, aura ripples, hair and sleeves move slightly, the NPC raises their gaze toward the viewer. One active NPC only; no subtitles, no readable text, no UI, no watermark, no logos, no black frame.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 4,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788025624377,
-    "promptEn": "cat-eared white-haired guard crouching in a ruined palace, purple aura around them",
-    "lastPlayedAt": 1788331024828,
-    "playCount": 22
-  },
-  {
-    "id": "b16b64e52f83",
-    "source": "h3console",
-    "prompt": "猫耳白发守卫蹲在废弃宫殿里，周身紫色气场",
-    "fullPrompt": "4-second image-to-video. Use the uploaded first frame as the exact opening composition: the current active NPC Neferpitou-like cat humanoid guard is clear in the foreground, white curly hair, cat ears, magenta eyes, dark outfit, crouched low with claws on the palace floor, violet aura behind. Fixed ruined palace room at night must remain consistent. Motion: subtle low camera push-in, aura ripples, hair and sleeves move slightly, the NPC raises their gaze toward the viewer. One active NPC only; no subtitles, no readable text, no UI, no watermark, no logos, no black frame.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 4,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788025623833,
-    "promptEn": "cat-eared white-haired guard crouching in a ruined palace, purple aura around them",
-    "lastPlayedAt": 1788326119314,
-    "playCount": 21
-  },
-  {
-    "id": "0d721acbc298",
-    "source": "h3console",
-    "prompt": "监控室里侦探眯眼看监视器",
-    "fullPrompt": "4-second image-to-video. Use the uploaded first frame as the exact opening composition: the current active NPC L-like detective is clearly visible crouched on a chair in a dark surveillance interrogation room, messy black hair, pale face, wide focused eyes, loose white shirt, knees drawn up, hand at chin. Fixed scene must remain the same: dim monitors, desk files, cool blue overhead light. Motion: slight camera push-in, monitor glow flickers, he tilts his head and narrows his eyes. One active NPC only; no subtitles, no readable text, no UI, no watermark, no logos, no black frame.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 4,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788025622363,
-    "promptEn": "detective in a surveillance room squinting at the monitors",
-    "lastPlayedAt": 1788327072124,
-    "playCount": 21
-  },
-  {
-    "id": "639069dea7c5",
-    "source": "h3console",
-    "prompt": "外星暴君举着能量球，站在绿色岩石战场冷笑",
-    "fullPrompt": "4-second image-to-video. Use the uploaded first frame as the exact opening composition: the current active NPC Frieza-like alien tyrant is large and clear in the foreground on a green rocky alien battlefield, purple-white armor plates, long tail, raised glowing energy sphere. Fixed scene and costume must remain consistent. Motion: subtle camera push-in, dust drifts, energy sphere pulses, the NPC slightly lowers his hand and sneers toward the viewer. One active NPC only; no subtitles, no readable text, no UI, no watermark, no logos, no black frame.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 4,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1788025620717,
-    "promptEn": "alien tyrant holding a glowing energy sphere, sneering on a rocky green battlefield",
-    "lastPlayedAt": 1788330554431,
-    "playCount": 22
-  },
-  {
-    "id": "9db41d807330",
-    "source": "h3console",
-    "prompt": "海盗船长把弯刀插在桌上，跟你谈危险的交易",
-    "fullPrompt": "4-second H3 text-to-video using the reference as identity and first-scene guidance. The first visible frame must clearly show the current active NPC, an older bearded cursed pirate captain in a large black tricorn hat and weathered coat, inside the Black Pearl captain cabin. Fixed scene: candlelit wooden captain table, gold coins, nautical charts, pistol on the table, moonlit porthole. Action: he plants a cutlass on the table and leans into a dangerous bargain, then lowers the blade slightly while watching the viewer. Keep one active NPC dominant, no subtitles, no UI, no watermark, no logos, no readable text.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 4,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788025220954,
-    "promptEn": "pirate captain plants a cutlass on the table, making a dangerous bargain",
-    "lastPlayedAt": 1788326231410,
-    "playCount": 21
-  },
-  {
-    "id": "3e03975ff40a",
-    "source": "h3console",
-    "prompt": "吸血鬼偶像红眼睛，在化妆间递出血红契约",
-    "fullPrompt": "4-second text-to-video. First frame must clearly show the current active NPC: an elegant vampire senior idol woman, occupying about 40 percent of the frame in the foreground, face, red eyes, stage costume, and hand holding a blood-red contract identifiable. Fixed scene: K-pop debut night backstage dressing room with lit vanity mirror and red stage light. Action: she leans close and offers the contract, subtle camera push-in. No subtitles, no readable text, no extra people, no UI, no watermark.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 4,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1788025003469,
-    "promptEn": "vampire idol with red eyes, offering a blood-red contract backstage",
-    "lastPlayedAt": 1788329565619,
-    "playCount": 21
-  },
-  {
-    "id": "2857aacf11bb",
-    "source": "h3console",
-    "prompt": "101号练习生站在高空玻璃桥上，挡路威胁你",
-    "fullPrompt": "4-second H3 text-to-video using the reference as identity and first-scene guidance. The first visible frame must clearly show the current active NPC, a stocky Korean survival-game bully in a green tracksuit marked 101, standing in the foreground on a high glass bridge arena. Fixed scene: alternating glass panels suspended high above darkness, colorful game-show walls, red countdown lights. Action: he blocks the path, leans forward and threatens the viewer, then points at the next glass panel. Keep one active NPC dominant, no subtitles, no UI, no watermark, no logos, no readable text except clothing number 101.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 4,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788024900966,
-    "promptEn": "player 101 blocking the path on a high glass bridge, threatening you",
-    "lastPlayedAt": 1788328989222,
-    "playCount": 21
-  },
-  {
-    "id": "d6aab3fce63d",
-    "source": "h3console",
-    "prompt": "受伤猎人靠着地铁闸机，让你往隧道走",
-    "fullPrompt": "4-second text-to-video. First frame must clearly show the current active NPC: a wounded S-rank hunter in modern black tactical armor, occupying about 40 percent of the frame in the foreground, face and bloodied bandage identifiable, leaning on a broken subway turnstile. Fixed scene: ruined subway dungeon entrance, blue lightning outside, dark tunnel behind. Action: he orders the viewer toward the tunnel while clutching his wound; tiny monster shadow only in background. No subtitles, no readable text, no extra people, no UI, no watermark.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 4,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1788024747209,
-    "promptEn": "wounded hunter leaning on a subway turnstile, ordering you into the tunnel",
-    "lastPlayedAt": 1788331104896,
-    "playCount": 22
-  },
-  {
-    "id": "fdefca4e4980",
-    "source": "h3console",
-    "prompt": "审判庭里太子签处决令，指向断头台",
-    "fullPrompt": "4-second text-to-video. First frame must clearly show the current active NPC: a cold young crown prince prosecutor in a dark royal court uniform, occupying about 40 percent of the frame in the foreground, face and upper body identifiable. Fixed scene: gothic royal trial hall and execution platform, candles, guillotine shadow, legal papers. Action: he presses an execution order on the desk and points toward the scaffold, stern expression, slight push-in camera. No subtitles, no readable text, no extra people, no UI, no watermark.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 4,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1788024491625,
-    "promptEn": "prince signs an execution order in the trial hall, points at the guillotine",
-    "lastPlayedAt": 1788330550428,
-    "playCount": 22
-  },
-  {
-    "id": "8a933feb52c4",
-    "source": "h3console",
-    "prompt": "捕快冲进门，回头看一眼，转身对屏风后女子说话",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with unnaturally pale skin, long black hair, and delicate, still features.\n<Subject 2> is the young male constable in <Picture 3>, wearing a dark indigo robe with a dao sword at his waist, a black head wrap, a waist tablet carved with the character \"捕\", and resolute, hard-set features.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens across the center, an unlit candle on the round stone table, a luohan daybed along the right wall, and a hanging wall scroll.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> bursting through the doorway into the room established by <Picture 1>, stumbling to a stop, glancing back at the open door, then turning toward the screens and speaking to <Subject 1>, who remains motionless behind the carved panels. The oil lamp flame reacts to the draft he brings in.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved screens, stone table, daybed, wall scroll, and dim single-source lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, pale complexion, and unnaturally still posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark indigo robe, dao sword, black head wrap, and waist tablet are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with one warm oil-lamp source against deep shadow and a slightly desaturated palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide view of the dim old room with five carved wooden screens standing across the center, an unlit candle resting on the round stone table, a luohan daybed along the right wall, and a hanging scroll on the back wall. A single oil lamp on a low table throws the only light in the room, its warm glow caught on the lacquered wood grain while the far corners stay buried in black. The wooden door bursts open at the right edge of frame and <Subject 2> stumbles inside, one hand catching the door frame as his boots slide across the dusty stone floor. The impact rattles the hanging scroll on the wall. A cold draft of night air reaches the lamp and the flame stretches sideways, dragging long, fast shadows over the carved openwork of the screens. <Subject 2> steadies himself with his chest heaving, breath showing faintly in the chill air, then twists his head back toward the open doorway and holds there, staring into the dark outside as if something is about to follow him in. Through the narrow vertical gaps between the screen panels, <Subject 1>, the woman in a flowing white robe with long black hair, stands perfectly motionless in the amber half-light, her stillness at odds with the commotion at the door.\n[Shot 2] At 00:03.500, the camera cuts to a medium close-up of <Subject 2> from a low angle, the lamp throwing upward light across his jaw and cheekbone while the carved screens fall into soft blur behind him. The camera pushes in with small amplitude at slow speed as he turns his head toward the screens, his right hand still braced on the door frame, and speaks in a tense, breathless voice, <d>[Chinese] 对不起，我不是有意闯进来的，外面有东西在追我。</d> His eyes stay fixed on the gap where <Subject 1>'s white sleeve is visible, and his knuckles whiten around the sword hilt without drawing it. He does not step further into the room. The lamp flame settles back into a slow sway, releasing the screens from the long shadows and leaving <Subject 1> a still, pale shape behind the carved wood while <Subject 2> holds his ground at the doorway, chest still rising and falling.\n\noverall_soundscape: The heavy wooden door slams open and rapid boots scrape across the stone floor. Night wind pushes through the doorway and the oil lamp flame hisses. Ragged breathing gradually slows, leaving the faint creak of old wood and the soft sputter of the wick.\n\nnon_diegetic_music: A low sustained guqin drone at a slow tempo, with a single plucked harmonic that decays before the final frame.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788019483999,
-    "promptEn": "constable rushes in, glances back at the door, then talks to the woman behind the screen",
-    "lastPlayedAt": 1788329573624,
-    "playCount": 21
-  },
-  {
-    "id": "998e3e6f9a2f",
-    "source": "h3console",
-    "prompt": "海盗谈判，抓住把柄反将巴博萨一军",
-    "fullPrompt": "Cinematic 真人影视 IP / 奇幻海盗电影 / 烛光船舱与诡计谈判, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制Hector Barbossa。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788016668421,
-    "promptEn": "pirate negotiation, catching Barbossa's flaw and turning the tables",
-    "lastPlayedAt": 1788331184954,
-    "playCount": 22
-  },
-  {
-    "id": "e3b18b5dfd8e",
-    "source": "h3console",
-    "prompt": "海盗谈判，抓住把柄反将巴博萨一军",
-    "fullPrompt": "Cinematic 真人影视 IP / 奇幻海盗电影 / 烛光船舱与诡计谈判, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制Hector Barbossa。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788016434846,
-    "promptEn": "pirate negotiation, catching Barbossa's flaw and turning the tables",
-    "lastPlayedAt": 1788329759772,
-    "playCount": 21
-  },
-  {
-    "id": "03e32ff5f0ef",
-    "source": "h3console",
-    "prompt": "海盗谈判，保住目标但代价不小",
-    "fullPrompt": "Cinematic 真人影视 IP / 奇幻海盗电影 / 烛光船舱与诡计谈判, an original high-stakes visual game scene, 带代价的局部成功 ending: 玩家守住核心目标但牺牲重要资源，留下下一关后果。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788016203093,
-    "promptEn": "pirate negotiation, held the goal but paid a price",
-    "lastPlayedAt": 1788328476830,
-    "playCount": 21
-  },
-  {
-    "id": "18aa882cbd0b",
-    "source": "h3console",
-    "prompt": "海盗谈判，保住目标但代价不小",
-    "fullPrompt": "Cinematic 真人影视 IP / 奇幻海盗电影 / 烛光船舱与诡计谈判, an original high-stakes visual game scene, 带代价的局部成功 ending: 玩家守住核心目标但牺牲重要资源，留下下一关后果。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788015968838,
-    "promptEn": "pirate negotiation, held the goal but paid a price",
-    "lastPlayedAt": 1788330446353,
-    "playCount": 22
-  },
-  {
-    "id": "b328649000d1",
-    "source": "h3console",
-    "prompt": "海盗谈判，正面硬刚被巴博萨看穿",
-    "fullPrompt": "Cinematic 真人影视 IP / 奇幻海盗电影 / 烛光船舱与诡计谈判, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被Hector Barbossa预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788015737873,
-    "promptEn": "pirate negotiation, went in head-on and Barbossa saw it coming",
-    "lastPlayedAt": 1788326109303,
-    "playCount": 21
-  },
-  {
-    "id": "2a5ee87b8f13",
-    "source": "h3console",
-    "prompt": "海盗谈判，正面硬刚被巴博萨看穿",
-    "fullPrompt": "Cinematic 真人影视 IP / 奇幻海盗电影 / 烛光船舱与诡计谈判, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被Hector Barbossa预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788015493445,
-    "promptEn": "pirate negotiation, went in head-on and Barbossa saw it coming",
-    "lastPlayedAt": 1788327096146,
-    "playCount": 21
-  },
-  {
-    "id": "3f30f5097438",
-    "source": "h3console",
-    "prompt": "海盗谈判，一直被动最后被逼到墙角",
-    "fullPrompt": "Cinematic 真人影视 IP / 奇幻海盗电影 / 烛光船舱与诡计谈判, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别Hector Barbossa的真实压迫，只是被动防守或沉默，威胁值达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788015035234,
-    "promptEn": "pirate negotiation, stayed passive and got cornered",
-    "lastPlayedAt": 1788328678983,
-    "playCount": 21
-  },
-  {
-    "id": "6a0d14cabe17",
-    "source": "h3console",
-    "prompt": "海盗谈判，一直被动最后被逼到墙角",
-    "fullPrompt": "Cinematic 真人影视 IP / 奇幻海盗电影 / 烛光船舱与诡计谈判, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别Hector Barbossa的真实压迫，只是被动防守或沉默，威胁值达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788014791733,
-    "promptEn": "pirate negotiation, stayed passive and got cornered",
-    "lastPlayedAt": 1788330472365,
-    "playCount": 22
-  },
-  {
-    "id": "08120928cc77",
-    "source": "h3console",
-    "prompt": "海盗谈判，一直被动最后被逼到墙角",
-    "fullPrompt": "Cinematic 真人影视 IP / 奇幻海盗电影 / 烛光船舱与诡计谈判, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别Hector Barbossa的真实压迫，只是被动防守或沉默，威胁值达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788014790969,
-    "promptEn": "pirate negotiation, stayed passive and got cornered",
-    "lastPlayedAt": 1788330540424,
-    "playCount": 22
-  },
-  {
-    "id": "e14fe7c3220e",
-    "source": "h3console",
-    "prompt": "黑珍珠号船长室，海盗包围，谈判随时可能变处刑",
-    "fullPrompt": "Cinematic 真人影视 IP / 奇幻海盗电影 / 烛光船舱与诡计谈判, 黑珍珠号甲板或船长室: 海盗包围、诅咒未解，谈判随时可能变成处刑。, opening tension between the player and Hector Barbossa, dramatic lighting, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788014563344,
-    "promptEn": "captain's cabin on the Black Pearl, surrounded by pirates, negotiation could turn into an execution",
-    "lastPlayedAt": 1788327034093,
-    "playCount": 21
-  },
-  {
-    "id": "c6e8560f62a2",
-    "source": "h3console",
-    "prompt": "黑珍珠号船长室，海盗包围，谈判随时可能变处刑",
-    "fullPrompt": "Cinematic 真人影视 IP / 奇幻海盗电影 / 烛光船舱与诡计谈判, 黑珍珠号甲板或船长室: 海盗包围、诅咒未解，谈判随时可能变成处刑。, opening tension between the player and Hector Barbossa, dramatic lighting, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788014336515,
-    "promptEn": "captain's cabin on the Black Pearl, surrounded by pirates, negotiation could turn into an execution",
-    "lastPlayedAt": 1788329569620,
-    "playCount": 21
-  },
-  {
-    "id": "eae105ee1204",
-    "source": "h3console",
-    "prompt": "玻璃桥游戏，抓准破绽逼退101号",
-    "fullPrompt": "Cinematic 真人影视 IP / 高饱和生存游戏剧 / 童趣空间与残酷规则反差, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制张德秀 Jang Deok-su / Player 101。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788014099497,
-    "promptEn": "glass bridge game, exploiting a flaw to push back player 101",
-    "lastPlayedAt": 1788330818652,
-    "playCount": 22
-  },
-  {
-    "id": "24576c264a25",
-    "source": "h3console",
-    "prompt": "玻璃桥游戏，抓准破绽逼退101号",
-    "fullPrompt": "Cinematic 真人影视 IP / 高饱和生存游戏剧 / 童趣空间与残酷规则反差, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制张德秀 Jang Deok-su / Player 101。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788013870849,
-    "promptEn": "glass bridge game, exploiting a flaw to push back player 101",
-    "lastPlayedAt": 1788329917915,
-    "playCount": 22
-  },
-  {
-    "id": "dd12bbb55060",
-    "source": "h3console",
-    "prompt": "玻璃桥游戏，保住目标但代价不小",
-    "fullPrompt": "Cinematic 真人影视 IP / 高饱和生存游戏剧 / 童趣空间与残酷规则反差, an original high-stakes visual game scene, 带代价的局部成功 ending: 玩家守住核心目标但牺牲重要资源，留下下一关后果。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788013634333,
-    "promptEn": "glass bridge game, held the goal but paid a price",
-    "lastPlayedAt": 1788330668524,
-    "playCount": 22
-  },
-  {
-    "id": "5b8ef21616d4",
-    "source": "h3console",
-    "prompt": "玻璃桥游戏，保住目标但代价不小",
-    "fullPrompt": "Cinematic 真人影视 IP / 高饱和生存游戏剧 / 童趣空间与残酷规则反差, an original high-stakes visual game scene, 带代价的局部成功 ending: 玩家守住核心目标但牺牲重要资源，留下下一关后果。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788013632065,
-    "promptEn": "glass bridge game, held the goal but paid a price",
-    "lastPlayedAt": 1788326257431,
-    "playCount": 21
-  },
-  {
-    "id": "892b923eed29",
-    "source": "h3console",
-    "prompt": "生存游戏里猜错对手想法，被反将一军",
-    "fullPrompt": "Cinematic 真人影视 IP / 高饱和生存游戏剧 / 童趣空间与残酷规则反差, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被张德秀 Jang Deok-su / Player 101预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788013393463,
-    "promptEn": "guessed wrong in a survival game, got countered",
-    "lastPlayedAt": 1788327106156,
-    "playCount": 21
-  },
-  {
-    "id": "7880a91928c6",
-    "source": "h3console",
-    "prompt": "玻璃桥游戏，一直被动最后被逼到墙角",
-    "fullPrompt": "Cinematic 真人影视 IP / 高饱和生存游戏剧 / 童趣空间与残酷规则反差, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别张德秀 Jang Deok-su / Player 101的真实压迫，只是被动防守或沉默，逼迫值达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788013155592,
-    "promptEn": "glass bridge game, stayed passive and got cornered",
-    "lastPlayedAt": 1788330122091,
-    "playCount": 22
-  },
-  {
-    "id": "f80c8ec1a8c8",
-    "source": "h3console",
-    "prompt": "站上玻璃桥，跟101号练习生对峙",
-    "fullPrompt": "Cinematic 真人影视 IP / 高饱和生存游戏剧 / 童趣空间与残酷规则反差, 玻璃桥游戏场: 高空、倒计时和真假玻璃让每一步都变成生死选择。, opening tension between the player and 张德秀 Jang Deok-su / Player 101, dramatic lighting, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788012917294,
-    "promptEn": "standing on the glass bridge, facing off with player 101",
-    "lastPlayedAt": 1788331433162,
-    "playCount": 22
-  },
-  {
-    "id": "8f23de61904f",
-    "source": "h3console",
-    "prompt": "站上玻璃桥，跟101号练习生对峙",
-    "fullPrompt": "Cinematic 真人影视 IP / 高饱和生存游戏剧 / 童趣空间与残酷规则反差, 玻璃桥游戏场: 高空、倒计时和真假玻璃让每一步都变成生死选择。, opening tension between the player and 张德秀 Jang Deok-su / Player 101, dramatic lighting, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788012917185,
-    "promptEn": "standing on the glass bridge, facing off with player 101",
-    "lastPlayedAt": 1788327326336,
-    "playCount": 21
-  },
-  {
-    "id": "bb4a8351d822",
-    "source": "h3console",
-    "prompt": "精神巢穴里，抓住破绽反制Vecna",
-    "fullPrompt": "Cinematic 真人影视 IP / 80 年代复古科幻恐怖 / 异世界精神空间, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制Vecna。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788012679740,
-    "promptEn": "in the mental lair, exploiting a flaw to counter Vecna",
-    "lastPlayedAt": 1788330272221,
-    "playCount": 22
-  },
-  {
-    "id": "3913752bbe3a",
-    "source": "h3console",
-    "prompt": "对抗Vecna，保住目标但代价不小",
-    "fullPrompt": "Cinematic 真人影视 IP / 80 年代复古科幻恐怖 / 异世界精神空间, an original high-stakes visual game scene, 带代价的局部成功 ending: 玩家守住核心目标但牺牲重要资源，留下下一关后果。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788012441458,
-    "promptEn": "facing Vecna, held the goal but paid a price",
-    "lastPlayedAt": 1788326199385,
-    "playCount": 21
-  },
-  {
-    "id": "691c2f679759",
-    "source": "h3console",
-    "prompt": "对抗Vecna，正面硬刚被看穿",
-    "fullPrompt": "Cinematic 真人影视 IP / 80 年代复古科幻恐怖 / 异世界精神空间, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被Vecna预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788012202802,
-    "promptEn": "facing Vecna, went in head-on and got seen through",
-    "lastPlayedAt": 1788326431578,
-    "playCount": 21
-  },
-  {
-    "id": "07f957783a09",
-    "source": "h3console",
-    "prompt": "对抗Vecna，正面硬刚被看穿",
-    "fullPrompt": "Cinematic 真人影视 IP / 80 年代复古科幻恐怖 / 异世界精神空间, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被Vecna预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788012202708,
-    "promptEn": "facing Vecna, went in head-on and got seen through",
-    "lastPlayedAt": 1788330828659,
-    "playCount": 22
-  },
-  {
-    "id": "40ccb04f76e9",
-    "source": "h3console",
-    "prompt": "对抗Vecna，一直被动最后被逼到墙角",
-    "fullPrompt": "Cinematic 真人影视 IP / 80 年代复古科幻恐怖 / 异世界精神空间, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别Vecna的真实压迫，只是被动防守或沉默，侵蚀值达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788011964828,
-    "promptEn": "facing Vecna, stayed passive and got cornered",
-    "lastPlayedAt": 1788327462455,
-    "playCount": 21
-  },
-  {
-    "id": "7161aecafa32",
-    "source": "h3console",
-    "prompt": "闯进Vecna的精神巢穴，红色闪电诡异空间",
-    "fullPrompt": "Cinematic 真人影视 IP / 80 年代复古科幻恐怖 / 异世界精神空间, Vecna 的精神巢穴: 扭曲的 Upside Down 空间、红色闪电和创伤记忆构成逃脱迷宫。, opening tension between the player and Vecna, dramatic lighting, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788011727444,
-    "promptEn": "entering Vecna's mental lair, red lightning, twisted space",
-    "lastPlayedAt": 1788326441585,
-    "playCount": 21
-  },
-  {
-    "id": "6a9af7e7368d",
-    "source": "h3console",
-    "prompt": "墓地对峙，抓住把柄反制伏地魔",
-    "fullPrompt": "Cinematic 真人影视 IP / 魔法奇幻电影 / 哥特墓地与黑魔法仪式, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制Voldemort。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788011488775,
-    "promptEn": "graveyard standoff, exploiting a flaw to counter Voldemort",
-    "lastPlayedAt": 1788330062036,
-    "playCount": 22
-  },
-  {
-    "id": "24845be71ea0",
-    "source": "h3console",
-    "prompt": "墓地对峙，抓住把柄反制伏地魔",
-    "fullPrompt": "Cinematic 真人影视 IP / 魔法奇幻电影 / 哥特墓地与黑魔法仪式, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制Voldemort。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788011488667,
-    "promptEn": "graveyard standoff, exploiting a flaw to counter Voldemort",
-    "lastPlayedAt": 1788330770616,
-    "playCount": 22
-  },
-  {
-    "id": "f1012a3d7aa7",
-    "source": "h3console",
-    "prompt": "对抗伏地魔，保住目标但代价不小",
-    "fullPrompt": "Cinematic 真人影视 IP / 魔法奇幻电影 / 哥特墓地与黑魔法仪式, an original high-stakes visual game scene, 带代价的局部成功 ending: 玩家守住核心目标但牺牲重要资源，留下下一关后果。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788011251158,
-    "promptEn": "facing Voldemort, held the goal but paid a price",
-    "lastPlayedAt": 1788330712564,
-    "playCount": 22
-  },
-  {
-    "id": "58f68b111429",
-    "source": "h3console",
-    "prompt": "对抗伏地魔，正面硬刚被看穿",
-    "fullPrompt": "Cinematic 真人影视 IP / 魔法奇幻电影 / 哥特墓地与黑魔法仪式, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被Voldemort预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788011013309,
-    "promptEn": "facing Voldemort, went in head-on and got seen through",
-    "lastPlayedAt": 1788330562441,
-    "playCount": 22
-  },
-  {
-    "id": "682cbc166479",
-    "source": "h3console",
-    "prompt": "对抗伏地魔，一直被动最后被逼到墙角",
-    "fullPrompt": "Cinematic 真人影视 IP / 魔法奇幻电影 / 哥特墓地与黑魔法仪式, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别Voldemort的真实压迫，只是被动防守或沉默，危险值达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788010775613,
-    "promptEn": "facing Voldemort, stayed passive and got cornered",
-    "lastPlayedAt": 1788330824656,
-    "playCount": 22
-  },
-  {
-    "id": "795842efa4a1",
-    "source": "h3console",
-    "prompt": "三强杯把你传送到墓地，伏地魔复活仪式刚完成",
-    "fullPrompt": "Cinematic 真人影视 IP / 魔法奇幻电影 / 哥特墓地与黑魔法仪式, 小汉格顿墓地: 三强争霸奖杯把玩家传送到墓地，伏地魔复活仪式已经完成。, opening tension between the player and Voldemort, dramatic lighting, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788010537514,
-    "promptEn": "the triwizard cup teleports you to the graveyard, Voldemort's revival just finished",
-    "lastPlayedAt": 1788329195322,
-    "playCount": 21
-  },
-  {
-    "id": "6ed9eaf4fa1d",
-    "source": "h3console",
-    "prompt": "对抗灭霸，抓住破绽反制",
-    "fullPrompt": "Cinematic 真人影视 IP / 好莱坞超级英雄电影 / 宇宙战争与末日史诗感, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制灭霸 Thanos。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788010299688,
-    "promptEn": "facing Thanos, exploiting a flaw to counter him",
-    "lastPlayedAt": 1788329673708,
-    "playCount": 21
-  },
-  {
-    "id": "469294a4ed09",
-    "source": "h3console",
-    "prompt": "对抗灭霸，保住目标但代价不小",
-    "fullPrompt": "Cinematic 真人影视 IP / 好莱坞超级英雄电影 / 宇宙战争与末日史诗感, an original high-stakes visual game scene, 带代价的局部成功 ending: 玩家守住核心目标但牺牲重要资源，留下下一关后果。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788010062535,
-    "promptEn": "facing Thanos, held the goal but paid a price",
-    "lastPlayedAt": 1788330242194,
-    "playCount": 22
-  },
-  {
-    "id": "324b84ebc607",
-    "source": "h3console",
-    "prompt": "对抗灭霸，正面硬刚被看穿",
-    "fullPrompt": "Cinematic 真人影视 IP / 好莱坞超级英雄电影 / 宇宙战争与末日史诗感, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被灭霸 Thanos预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788009825082,
-    "promptEn": "facing Thanos, went in head-on and got seen through",
-    "lastPlayedAt": 1788330572447,
-    "playCount": 22
-  },
-  {
-    "id": "a6ead2cbe03f",
-    "source": "h3console",
-    "prompt": "对抗灭霸，一直被动最后被逼到墙角",
-    "fullPrompt": "Cinematic 真人影视 IP / 好莱坞超级英雄电影 / 宇宙战争与末日史诗感, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别灭霸 Thanos的真实压迫，只是被动防守或沉默，压迫值达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788009586293,
-    "promptEn": "facing Thanos, stayed passive and got cornered",
-    "lastPlayedAt": 1788330476371,
-    "playCount": 22
-  },
-  {
-    "id": "438c98cf4973",
-    "source": "h3console",
-    "prompt": "泰坦星战场，最后阻止灭霸的机会",
-    "fullPrompt": "Cinematic 真人影视 IP / 好莱坞超级英雄电影 / 宇宙战争与末日史诗感, 泰坦星战场: 废墟、飞船残骸和无限手套的光芒构成最后阻止灭霸的机会。, opening tension between the player and 灭霸 Thanos, dramatic lighting, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788009346020,
-    "promptEn": "battlefield on Titan, the last chance to stop Thanos",
-    "lastPlayedAt": 1788329001230,
-    "playCount": 21
-  },
-  {
-    "id": "66a96c6bc7cb",
-    "source": "h3console",
-    "prompt": "无限旅馆里，抓住破绽反制永恒恶魔",
-    "fullPrompt": "Cinematic 动漫 IP / 血腥荒诞都市番 / 封闭空间循环惊悚, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制永恒恶魔 Eternity Devil。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788009108302,
-    "promptEn": "in the endless hotel, exploiting a flaw to counter the Eternity Devil",
-    "lastPlayedAt": 1788328658969,
-    "playCount": 21
-  },
-  {
-    "id": "71c590b0ea8f",
-    "source": "h3console",
-    "prompt": "对抗永恒恶魔，保住目标但代价不小",
-    "fullPrompt": "Cinematic 动漫 IP / 血腥荒诞都市番 / 封闭空间循环惊悚, an original high-stakes visual game scene, 带代价的局部成功 ending: 玩家守住核心目标但牺牲重要资源，留下下一关后果。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788008870118,
-    "promptEn": "facing the Eternity Devil, held the goal but paid a price",
-    "lastPlayedAt": 1788327734647,
-    "playCount": 21
-  },
-  {
-    "id": "35833c2cabb9",
-    "source": "h3console",
-    "prompt": "对抗永恒恶魔，正面硬刚被看穿",
-    "fullPrompt": "Cinematic 动漫 IP / 血腥荒诞都市番 / 封闭空间循环惊悚, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被永恒恶魔 Eternity Devil预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788008632603,
-    "promptEn": "facing the Eternity Devil, went in head-on and got seen through",
-    "lastPlayedAt": 1788330984794,
-    "playCount": 22
-  },
-  {
-    "id": "361c80537652",
-    "source": "h3console",
-    "prompt": "对抗永恒恶魔，一直被动最后被逼到墙角",
-    "fullPrompt": "Cinematic 动漫 IP / 血腥荒诞都市番 / 封闭空间循环惊悚, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别永恒恶魔 Eternity Devil的真实压迫，只是被动防守或沉默，侵蚀值达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788008394344,
-    "promptEn": "facing the Eternity Devil, stayed passive and got cornered",
-    "lastPlayedAt": 1788330576448,
-    "playCount": 22
-  },
-  {
-    "id": "1a57de01bad0",
-    "source": "h3console",
-    "prompt": "被永恒恶魔封锁的酒店楼层，走廊无限循环",
-    "fullPrompt": "Cinematic 动漫 IP / 血腥荒诞都市番 / 封闭空间循环惊悚, 被永恒恶魔封锁的酒店楼层: 走廊无限循环，同伴开始崩溃。, opening tension between the player and 永恒恶魔 Eternity Devil, dramatic lighting, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788008156307,
-    "promptEn": "hotel floor sealed by the Eternity Devil, an endless looping corridor",
-    "lastPlayedAt": 1788330162130,
-    "playCount": 22
-  },
-  {
-    "id": "cd3b6dda90ad",
-    "source": "h3console",
-    "prompt": "城市废墟里，抓住破绽反制死柄木弔",
-    "fullPrompt": "Cinematic 动漫 IP / 英雄漫画战斗番 / 都市灾难与英雄信念, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制死柄木弔 Tomura Shigaraki。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788007915610,
-    "promptEn": "in the ruined city, exploiting a flaw to counter Tomura",
-    "lastPlayedAt": 1788331295047,
-    "playCount": 22
-  },
-  {
-    "id": "f67685cd6862",
-    "source": "h3console",
-    "prompt": "对抗死柄木弔，保住目标但代价不小",
-    "fullPrompt": "Cinematic 动漫 IP / 英雄漫画战斗番 / 都市灾难与英雄信念, an original high-stakes visual game scene, 带代价的局部成功 ending: 玩家守住核心目标但牺牲重要资源，留下下一关后果。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788007678723,
-    "promptEn": "facing Tomura, held the goal but paid a price",
-    "lastPlayedAt": 1788330688541,
-    "playCount": 22
-  },
-  {
-    "id": "ee5bd111d8bd",
-    "source": "h3console",
-    "prompt": "神野区战场，城市被毁，英雄陷入危机",
-    "fullPrompt": "Cinematic 动漫 IP / 英雄漫画战斗番 / 都市灾难与英雄信念, 神野区战斗现场: 城市被破坏，爆豪被困，反派联盟试图动摇英雄社会。, opening tension between the player and 死柄木弔 Tomura Shigaraki, dramatic lighting, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788007441156,
-    "promptEn": "battlefield in Jaku city, the city in ruins, heroes in crisis",
-    "lastPlayedAt": 1788330582456,
-    "playCount": 22
-  },
-  {
-    "id": "68219d6b68c8",
-    "source": "h3console",
-    "prompt": "闯进王宫，尼飞彼多守着蚁王拖延时间",
-    "fullPrompt": "Cinematic 动漫 IP / 黑暗奇幻战斗番 / 王宫压迫与生死谈判, 东果陀王宫突入区: 王宫寂静但危机极高，尼飞彼多守护蚁王并拖延时间。, opening tension between the player and 尼飞彼多 Neferpitou, dramatic lighting, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788007201908,
-    "promptEn": "breaking into the palace, Neferpitou stalling to protect the king",
-    "lastPlayedAt": 1788326335495,
-    "playCount": 21
-  },
-  {
-    "id": "628026375795",
-    "source": "h3console",
-    "prompt": "抓住证据破绽，反制L",
-    "fullPrompt": "Cinematic 动漫 IP / 悬疑心理战 / 冷色犯罪推理感, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制L Lawliet。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788006964122,
-    "promptEn": "grabbing an evidence flaw to counter L",
-    "lastPlayedAt": 1788330356289,
-    "playCount": 22
-  },
-  {
-    "id": "0c6306b07f79",
-    "source": "h3console",
-    "prompt": "跟L交锋，正面硬刚被看穿",
-    "fullPrompt": "Cinematic 动漫 IP / 悬疑心理战 / 冷色犯罪推理感, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被L Lawliet预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788006725887,
-    "promptEn": "facing off with L, went in head-on and got seen through",
-    "lastPlayedAt": 1788330234186,
-    "playCount": 22
-  },
-  {
-    "id": "ee82188227fa",
-    "source": "h3console",
-    "prompt": "跟L对峙，一直被动最后证据压力爆表",
-    "fullPrompt": "Cinematic 动漫 IP / 悬疑心理战 / 冷色犯罪推理感, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别L Lawliet的真实压迫，只是被动防守或沉默，证据压力达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788006252918,
-    "promptEn": "facing L, stayed passive and the evidence pressure maxed out",
-    "lastPlayedAt": 1788328472824,
-    "playCount": 21
-  },
-  {
-    "id": "0e28a7fecf03",
-    "source": "h3console",
-    "prompt": "跟L对峙，一直被动最后证据压力爆表",
-    "fullPrompt": "Cinematic 动漫 IP / 悬疑心理战 / 冷色犯罪推理感, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别L Lawliet的真实压迫，只是被动防守或沉默，证据压力达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788006252824,
-    "promptEn": "facing L, stayed passive and the evidence pressure maxed out",
-    "lastPlayedAt": 1788327556528,
-    "playCount": 21
-  },
-  {
-    "id": "fb0738ad3e3b",
-    "source": "h3console",
-    "prompt": "荒星战场，抓住破绽反制弗利萨",
-    "fullPrompt": "Cinematic 动漫 IP / 高能战斗番 / 荒星地貌与能量爆发, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制弗利萨 Frieza。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788006015175,
-    "promptEn": "on the ruined planet, exploiting a flaw to counter Frieza",
-    "lastPlayedAt": 1788329559616,
-    "playCount": 21
-  },
-  {
-    "id": "e6fda5c2237d",
-    "source": "h3console",
-    "prompt": "对抗弗利萨，保住目标但代价不小",
-    "fullPrompt": "Cinematic 动漫 IP / 高能战斗番 / 荒星地貌与能量爆发, an original high-stakes visual game scene, 带代价的局部成功 ending: 玩家守住核心目标但牺牲重要资源，留下下一关后果。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788005778588,
-    "promptEn": "facing Frieza, held the goal but paid a price",
-    "lastPlayedAt": 1788330466362,
-    "playCount": 22
-  },
-  {
-    "id": "b96cc9762cf6",
-    "source": "h3console",
-    "prompt": "对抗弗利萨，正面硬刚被看穿",
-    "fullPrompt": "Cinematic 动漫 IP / 高能战斗番 / 荒星地貌与能量爆发, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被弗利萨 Frieza预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788005541640,
-    "promptEn": "facing Frieza, went in head-on and got seen through",
-    "lastPlayedAt": 1788328993225,
-    "playCount": 21
-  },
-  {
-    "id": "574b47b00316",
-    "source": "h3console",
-    "prompt": "对抗弗利萨，一直被动最后被逼到墙角",
-    "fullPrompt": "Cinematic 动漫 IP / 高能战斗番 / 荒星地貌与能量爆发, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别弗利萨 Frieza的真实压迫，只是被动防守或沉默，压迫值达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788005305091,
-    "promptEn": "facing Frieza, stayed passive and got cornered",
-    "lastPlayedAt": 1788330544429,
-    "playCount": 22
-  },
-  {
-    "id": "d680e111ade5",
-    "source": "h3console",
-    "prompt": "那美克星要毁灭了，龙珠愿望还在争夺",
-    "fullPrompt": "Cinematic 动漫 IP / 高能战斗番 / 荒星地貌与能量爆发, 崩坏中的那美克星地表: 龙珠愿望即将被争夺，星球随时可能毁灭。, opening tension between the player and 弗利萨 Frieza, dramatic lighting, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788005066484,
-    "promptEn": "Namek is about to be destroyed, the dragon ball wish still up for grabs",
-    "lastPlayedAt": 1788327366373,
-    "playCount": 21
-  },
-  {
-    "id": "565059763888",
-    "source": "h3console",
-    "prompt": "地铁站里，抓住破绽反制真人",
-    "fullPrompt": "Cinematic 动漫 IP / 都市黑暗战斗番 / 地铁封锁与咒灵压迫, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制真人 Mahito。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788004828526,
-    "promptEn": "in the subway, exploiting a flaw to counter Mahito",
-    "lastPlayedAt": 1788326063276,
-    "playCount": 21
-  },
-  {
-    "id": "323737948733",
-    "source": "h3console",
-    "prompt": "对抗真人，保住目标但代价不小",
-    "fullPrompt": "Cinematic 动漫 IP / 都市黑暗战斗番 / 地铁封锁与咒灵压迫, an original high-stakes visual game scene, 带代价的局部成功 ending: 玩家守住核心目标但牺牲重要资源，留下下一关后果。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788004590937,
-    "promptEn": "facing Mahito, held the goal but paid a price",
-    "lastPlayedAt": 1788331457184,
-    "playCount": 22
-  },
-  {
-    "id": "9996a8ac8326",
-    "source": "h3console",
-    "prompt": "对抗真人，正面硬刚被看穿",
-    "fullPrompt": "Cinematic 动漫 IP / 都市黑暗战斗番 / 地铁封锁与咒灵压迫, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被真人 Mahito预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788004353928,
-    "promptEn": "facing Mahito, went in head-on and got seen through",
-    "lastPlayedAt": 1788328741036,
-    "playCount": 21
-  },
-  {
-    "id": "971d80ec087c",
-    "source": "h3console",
-    "prompt": "对抗真人，一直被动最后被逼到墙角",
-    "fullPrompt": "Cinematic 动漫 IP / 都市黑暗战斗番 / 地铁封锁与咒灵压迫, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别真人 Mahito的真实压迫，只是被动防守或沉默，压迫值达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788004115725,
-    "promptEn": "facing Mahito, stayed passive and got cornered",
-    "lastPlayedAt": 1788331166940,
-    "playCount": 22
-  },
-  {
-    "id": "5dd52c568f66",
-    "source": "h3console",
-    "prompt": "对抗真人，一直被动最后被逼到墙角",
-    "fullPrompt": "Cinematic 动漫 IP / 都市黑暗战斗番 / 地铁封锁与咒灵压迫, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别真人 Mahito的真实压迫，只是被动防守或沉默，压迫值达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788004115612,
-    "promptEn": "facing Mahito, stayed passive and got cornered",
-    "lastPlayedAt": 1788330834662,
-    "playCount": 22
-  },
-  {
-    "id": "f4a2870386c6",
-    "source": "h3console",
-    "prompt": "梦境里，抓住破绽反制魇梦",
-    "fullPrompt": "Cinematic 动漫 IP / 日式奇幻战斗番 / 梦境惊悚与华丽刀光, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制魇梦 Enmu。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788003877425,
-    "promptEn": "in the dream, exploiting a flaw to counter Enmu",
-    "lastPlayedAt": 1788330734588,
-    "playCount": 22
-  },
-  {
-    "id": "7c4f9460e6b7",
-    "source": "h3console",
-    "prompt": "对抗魇梦，保住目标但代价不小",
-    "fullPrompt": "Cinematic 动漫 IP / 日式奇幻战斗番 / 梦境惊悚与华丽刀光, an original high-stakes visual game scene, 带代价的局部成功 ending: 玩家守住核心目标但牺牲重要资源，留下下一关后果。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788003639415,
-    "promptEn": "facing Enmu, held the goal but paid a price",
-    "lastPlayedAt": 1788330786628,
-    "playCount": 22
-  },
-  {
-    "id": "6881dbe839a0",
-    "source": "h3console",
-    "prompt": "对抗魇梦，正面硬刚被看穿",
-    "fullPrompt": "Cinematic 动漫 IP / 日式奇幻战斗番 / 梦境惊悚与华丽刀光, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被魇梦 Enmu预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788003402559,
-    "promptEn": "facing Enmu, went in head-on and got seen through",
-    "lastPlayedAt": 1788326861945,
-    "playCount": 21
-  },
-  {
-    "id": "5c0f1663f4ae",
-    "source": "h3console",
-    "prompt": "对抗魇梦，一直被动最后被逼到墙角",
-    "fullPrompt": "Cinematic 动漫 IP / 日式奇幻战斗番 / 梦境惊悚与华丽刀光, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别魇梦 Enmu的真实压迫，只是被动防守或沉默，侵蚀值达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788003166117,
-    "promptEn": "facing Enmu, stayed passive and got cornered",
-    "lastPlayedAt": 1788330534418,
-    "playCount": 22
-  },
-  {
-    "id": "a3f39da74c1f",
-    "source": "h3console",
-    "prompt": "对抗莱纳，抓住破绽反制",
-    "fullPrompt": "Cinematic 动漫 IP / 冷峻写实战争番 / 末日压迫感, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制莱纳 Reiner Braun。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788002929455,
-    "promptEn": "facing Reiner, exploiting a flaw to counter him",
-    "lastPlayedAt": 1788328262637,
-    "playCount": 21
-  },
-  {
-    "id": "ea8b38014674",
-    "source": "h3console",
-    "prompt": "对抗莱纳，保住目标但代价不小",
-    "fullPrompt": "Cinematic 动漫 IP / 冷峻写实战争番 / 末日压迫感, an original high-stakes visual game scene, 带代价的局部成功 ending: 玩家守住核心目标但牺牲重要资源，留下下一关后果。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788002691822,
-    "promptEn": "facing Reiner, held the goal but paid a price",
-    "lastPlayedAt": 1788329817825,
-    "playCount": 22
-  },
-  {
-    "id": "f65b15d67966",
-    "source": "h3console",
-    "prompt": "对抗莱纳，正面硬刚被看穿",
-    "fullPrompt": "Cinematic 动漫 IP / 冷峻写实战争番 / 末日压迫感, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被莱纳 Reiner Braun预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788002455066,
-    "promptEn": "facing Reiner, went in head-on and got seen through",
-    "lastPlayedAt": 1788329841848,
-    "playCount": 22
-  },
-  {
-    "id": "5846d645e870",
-    "source": "h3console",
-    "prompt": "对抗莱纳，一直被动最后被逼到墙角",
-    "fullPrompt": "Cinematic 动漫 IP / 冷峻写实战争番 / 末日压迫感, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别莱纳 Reiner Braun的真实压迫，只是被动防守或沉默，压迫值达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788002217928,
-    "promptEn": "facing Reiner, stayed passive and got cornered",
-    "lastPlayedAt": 1788329811821,
-    "playCount": 22
-  },
-  {
-    "id": "3fedf5531999",
-    "source": "h3console",
-    "prompt": "希干希纳地下室，巨人的秘密压在肩上",
-    "fullPrompt": "Cinematic 动漫 IP / 冷峻写实战争番 / 末日压迫感, 希干希纳区地下室入口附近: 城镇废墟、巨人威胁和世界真相都压在玩家面前。, opening tension between the player and 莱纳 Reiner Braun, dramatic lighting, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788001981272,
-    "promptEn": "basement in Shiganshina, the titan secret weighing down on you",
-    "lastPlayedAt": 1788326147338,
-    "playCount": 21
-  },
-  {
-    "id": "7c2351a27a96",
-    "source": "h3console",
-    "prompt": "木叶废墟里，抓住破绽反制佩恩",
-    "fullPrompt": "Cinematic 动漫 IP / 忍者战斗番 / 废墟末日与情绪爆发, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制佩恩 Deva Path。, fixed main scene atmosphere, no subtitles, no UI, no watermark. Voice/dialogue rule: if the video script includes any human voice, speech, dialogue, chanting, narration, or vocal line, write it in English only; do not include Chinese spoken dialogue. Keep subtitles/UI/text absent. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788001508988,
-    "promptEn": "in the ruins of Konoha, exploiting a flaw to counter Pain",
-    "lastPlayedAt": 1788330660531,
-    "playCount": 22
-  },
-  {
-    "id": "79304c061606",
-    "source": "h3console",
-    "prompt": "木叶废墟里，抓住破绽反制佩恩",
-    "fullPrompt": "Cinematic 动漫 IP / 忍者战斗番 / 废墟末日与情绪爆发, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制佩恩 Deva Path。, fixed main scene atmosphere, no subtitles, no UI, no watermark. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness. If any human voice, speech, dialogue, chanting, narration, or vocal line appears, it must be in English only; no Chinese spoken dialogue.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788001428705,
-    "promptEn": "in the ruins of Konoha, exploiting a flaw to counter Pain",
-    "lastPlayedAt": 1788326151340,
-    "playCount": 21
-  },
-  {
-    "id": "e486fab4c24d",
-    "source": "h3console",
-    "prompt": "木叶废墟里，抓住破绽反制佩恩",
-    "fullPrompt": "Cinematic 动漫 IP / 忍者战斗番 / 废墟末日与情绪爆发, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制佩恩 Deva Path。, fixed main scene atmosphere, no subtitles, no UI, no watermark. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788001285796,
-    "promptEn": "in the ruins of Konoha, exploiting a flaw to counter Pain",
-    "lastPlayedAt": 1788329283385,
-    "playCount": 21
-  },
-  {
-    "id": "808e89ca06a4",
-    "source": "h3console",
-    "prompt": "对抗佩恩，保住目标但代价不小",
-    "fullPrompt": "Cinematic 动漫 IP / 忍者战斗番 / 废墟末日与情绪爆发, an original high-stakes visual game scene, 带代价的局部成功 ending: 玩家守住核心目标但牺牲重要资源，留下下一关后果。, fixed main scene atmosphere, no subtitles, no UI, no watermark. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788001049233,
-    "promptEn": "facing Pain, held the goal but paid a price",
-    "lastPlayedAt": 1788331084876,
-    "playCount": 22
-  },
-  {
-    "id": "7ecd480a3c31",
-    "source": "h3console",
-    "prompt": "对抗佩恩，正面硬刚被看穿",
-    "fullPrompt": "Cinematic 动漫 IP / 忍者战斗番 / 废墟末日与情绪爆发, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被佩恩 Deva Path预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788000812932,
-    "promptEn": "facing Pain, went in head-on and got seen through",
-    "lastPlayedAt": 1788329821833,
-    "playCount": 22
-  },
-  {
-    "id": "f691ac021df0",
-    "source": "h3console",
-    "prompt": "对抗佩恩，一直被动最后被逼到墙角",
-    "fullPrompt": "Cinematic 动漫 IP / 忍者战斗番 / 废墟末日与情绪爆发, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别佩恩 Deva Path的真实压迫，只是被动防守或沉默，压迫值达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788000577474,
-    "promptEn": "facing Pain, stayed passive and got cornered",
-    "lastPlayedAt": 1788330896716,
-    "playCount": 22
-  },
-  {
-    "id": "5ef9489bdc5b",
-    "source": "h3console",
-    "prompt": "木叶被神罗天征摧毁，佩恩站在废墟宣告秩序",
-    "fullPrompt": "Cinematic 动漫 IP / 忍者战斗番 / 废墟末日与情绪爆发, 被神罗天征摧毁后的木叶废墟: 村民受伤，佩恩站在废墟中宣告痛苦的秩序。, opening tension between the player and 佩恩 Deva Path, dramatic lighting, no subtitles, no UI, no watermark. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788000341927,
-    "promptEn": "Konoha destroyed by Shinra Tensei, Pain declaring order in the ruins",
-    "lastPlayedAt": 1788330764618,
-    "playCount": 22
-  },
-  {
-    "id": "e245dbcc6d36",
-    "source": "h3console",
-    "prompt": "对抗赤犬，抓住破绽反制",
-    "fullPrompt": "Cinematic 动漫 IP / 日式热血少年漫 / 大规模战争史诗感, an original high-stakes visual game scene, 最佳反制 ending: 玩家收集关键证据/破绽，利用场景资源反过来限制赤犬 Sakazuki。, fixed main scene atmosphere, no subtitles, no UI, no watermark. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1788000105889,
-    "promptEn": "facing Sakazuki, exploiting a flaw to counter him",
-    "lastPlayedAt": 1788327090140,
-    "playCount": 21
-  },
-  {
-    "id": "c9be7c07b793",
-    "source": "h3console",
-    "prompt": "对抗赤犬，保住目标但代价不小",
-    "fullPrompt": "Cinematic 动漫 IP / 日式热血少年漫 / 大规模战争史诗感, an original high-stakes visual game scene, 带代价的局部成功 ending: 玩家守住核心目标但牺牲重要资源，留下下一关后果。, fixed main scene atmosphere, no subtitles, no UI, no watermark. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787999869694,
-    "promptEn": "facing Sakazuki, held the goal but paid a price",
-    "lastPlayedAt": 1788331475202,
-    "playCount": 22
-  },
-  {
-    "id": "677941e683c4",
-    "source": "h3console",
-    "prompt": "对抗赤犬，正面硬刚被看穿",
-    "fullPrompt": "Cinematic 动漫 IP / 日式热血少年漫 / 大规模战争史诗感, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被赤犬 Sakazuki预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787999633745,
-    "promptEn": "facing Sakazuki, went in head-on and got seen through",
-    "lastPlayedAt": 1788327846733,
-    "playCount": 21
-  },
-  {
-    "id": "932d15f7f8bd",
-    "source": "h3console",
-    "prompt": "对抗赤犬，正面硬刚被看穿",
-    "fullPrompt": "Cinematic 动漫 IP / 日式热血少年漫 / 大规模战争史诗感, an original high-stakes visual game scene, 错误强攻/错误让步 ending: 玩家用最直观但被赤犬 Sakazuki预判的方式行动，触发反制。, fixed main scene atmosphere, no subtitles, no UI, no watermark. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787999633670,
-    "promptEn": "facing Sakazuki, went in head-on and got seen through",
-    "lastPlayedAt": 1788329181315,
-    "playCount": 21
-  },
-  {
-    "id": "a8b8b4430ee3",
-    "source": "h3console",
-    "prompt": "对抗赤犬，一直被动最后被逼到墙角",
-    "fullPrompt": "Cinematic 动漫 IP / 日式热血少年漫 / 大规模战争史诗感, an original high-stakes visual game scene, 拖延失败 ending: 玩家没有识别赤犬 Sakazuki的真实压迫，只是被动防守或沉默，压迫值达到上限。, fixed main scene atmosphere, no subtitles, no UI, no watermark. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787999397097,
-    "promptEn": "facing Sakazuki, stayed passive and got cornered",
-    "lastPlayedAt": 1788328963208,
-    "playCount": 21
-  },
-  {
-    "id": "88923aaf57f3",
-    "source": "h3console",
-    "prompt": "马林梵多刑场，艾斯要被处决，海军对峙海贼团",
-    "fullPrompt": "Cinematic 动漫 IP / 日式热血少年漫 / 大规模战争史诗感, 马林梵多处刑台附近的主战场: Ace 即将被处刑，海军、白胡子海贼团和路飞在同一战场上爆发冲突。, opening tension between the player and 赤犬 Sakazuki, dramatic lighting, no subtitles, no UI, no watermark. H3 fallback generation, cinematic motion, no subtitles, no UI, no watermark, avoid recognizable copyrighted logos or exact franchise character likeness.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787999161013,
-    "promptEn": "execution platform at Marineford, Ace about to be executed, navy vs pirates",
-    "lastPlayedAt": 1788330146111,
-    "playCount": 22
-  },
-  {
-    "id": "00476b62bf75",
-    "source": "h3console",
-    "prompt": "捕快问女子是不是一直一个人在这，她不说话",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens standing at its center, a round stone table holding an unlit candle, a luohan bed against the right wall, and a hanging scroll on the wall.\n<Subject 1> is the mysterious woman in flowing white ancient robes with a delicate, unnaturally pale face and long black hair, as established in <Picture 2>.\n<Subject 2> is the ancient constable in a dark robe with a dao sword at his waist, a black head cloth, black boots, and a waist tag bearing the Chinese character 捕, with a stern, resolute face, as established in <Picture 3>.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing before the carved screens in the room established by <Picture 1> and asking <Subject 1> whether she has always been alone in this place. The camera pushes in as the question hangs in the lamplit air, then cuts to <Subject 1> behind the screen gaps, who neither moves nor answers within the video.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, the five carved screens, the round stone table with the unlit candle, the luohan bed, the wall scroll, and the oil-lamp lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, delicate pale complexion, and still bearing are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark robe, dao sword, waist tag, black head cloth, and black boots are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with a single warm oil-lamp source against deep surrounding shadow and a slightly desaturated palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. The oil lamp on the low table burns as the only lit flame, its amber glow reaching the near edge of the carved screens, while the candle on the round stone table stands unlit among cold wax. <Subject 2>, the constable in a dark robe with a dao sword at his waist and a black head cloth, stands in the foreground facing the screens, his weight planted evenly on the stone floor. The camera pushes in with small amplitude at slow speed toward the narrow gaps between the screen panels. <Subject 2> (S1) speaks in a low, measured voice with a careful, unhurried pace, <d>[Chinese] 那你呢，你一直都一个人在这里？</d> His right hand rests near the dao hilt without drawing, and his eyes stay fixed on the gaps where a pale sleeve is faintly visible behind the carved wood.\n[Shot 2] At 00:03.500, the camera cuts to a close-up framed through a narrow gap between two screen panels on <Subject 1>, the woman in flowing white robes with long black hair and a delicate, unnaturally pale face. The carved wood borders her figure on both sides, and the amber lamp light catches one side of her face while the rest of her stays in soft shadow. The camera holds a static shot. Her expression remains neutral and unreadable as her gaze drifts slowly toward the direction of his voice, and she does not speak. The oil lamp flame sways once, deepening the shadow across the carved screen surface before settling, and the question lingers unanswered through the end of the video.\n\noverall_soundscape: The oil lamp wick emits a faint continuous sputter, and the old wooden screen frame creaks softly once. A subtle rustle of silk shifts behind the panels, and distant wind presses against the exterior wall.\n\nnon_diegetic_music: A sustained low erhu drone at a very slow tempo, joined by a single sparse guzheng harmonic that fades without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787903583872,
-    "promptEn": "constable asks if she's always been alone here, she doesn't answer",
-    "lastPlayedAt": 1788329631672,
-    "playCount": 21
-  },
-  {
-    "id": "2ab2094e5679",
-    "source": "h3console",
-    "prompt": "捕快不喝酒，问这是什么地方，女子隔着屏风看他",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with a delicate, unnaturally pale face, long black hair, and an unnaturally still posture.\n<Subject 2> is a young male constable in a dark long robe with a dao sword at his waist, a black headwrap, a waist plaque engraved with \"捕\", and a resolute face, standing in the room established by <Picture 1>.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens standing across the center, an oil lamp on the round stone table, an unlit candle beside it, a luohan couch along the right wall, and a hanging scroll on the wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing in the lamp-lit room established by <Picture 1>, speaking toward <Subject 1> behind the carved screens: he declines to drink and asks her to tell him what this place is. The camera pushes in slowly on <Subject 2>, then cuts to <Subject 1>'s pale face watching him through the screen gaps. Neither figure changes position and she gives no reply.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved screens, oil lamp on the round stone table, unlit candle, luohan couch, and hanging scroll are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, pale complexion, and still posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark long robe, dao sword, black headwrap, and waist plaque are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with a single warm amber source from the oil lamp against deep shadow and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens standing across the center, the oil lamp burning low on the round stone table, and the candle beside it standing unlit with no flame. <Subject 2>, the constable in a dark long robe with a dao sword at his waist and a black headwrap, stands in the foreground facing the screens, his weight evenly planted. The camera pushes in with small amplitude at slow speed toward him. In a low, steady male voice with a measured pace, <Subject 2> (S1) says: <d>[Chinese] 我不渴，你先告诉我这里是什么地方。</d> Through the narrow gaps between the screen panels, the faint silhouette of <Subject 1>, the woman in a white robe with long black hair and a pale, delicate face, remains motionless with her lips closed. The lamp flame sways gently, shifting amber light across the carved screen surfaces.\n[Shot 2] At 00:03.500, the camera cuts to a close-up of <Subject 1>'s pale face seen through a narrow gap between two screen panels, the carved wood framing her gaze on both sides. The amber lamp light catches the unnatural pallor of her cheek while her dark eyes stay fixed on <Subject 2>, her expression neutral and unreadable, neither inviting nor refusing. The camera holds a static shot. A single strand of black hair shifts slightly across her cheekbone from an imperceptible air current, and the oil lamp flame settles back to a steady glow. She gives no reply, and the silence holds to the end of the shot.\n\noverall_soundscape: Near-silence. The oil lamp wick emits a faint continuous sputter. A subtle rustle of silk comes from behind the screens once, and distant wind presses against the exterior wall.\n\nnon_diegetic_music: A sustained low erhu drone at a very slow tempo, joined by a single sparse pipa pluck that fades without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787900324972,
-    "promptEn": "constable declines a drink, asks what this place is, she watches him through the screen",
-    "lastPlayedAt": 1788330380310,
-    "playCount": 22
-  },
-  {
-    "id": "43ee5c728f1b",
-    "source": "h3console",
-    "prompt": "捕快伸手想碰女子的手，隔着屏风缝隙，她没反应",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with an exquisitely delicate face of unnaturally pale complexion and long black hair.\n<Subject 2> is the young male constable in <Picture 3>, wearing a dark robe, a black head wrap, and black boots, with a resolute jaw, a waist tag bearing the visible character \"捕\" at his belt, and a dao sword at his waist.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens standing in the center, a round stone table carrying an unlit candle, a luohan couch on the right, and a hanging scroll on the wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing before the carved screens in the room established by <Picture 1>, extending one hand palm-up toward the narrow gap where <Subject 1> is partially visible, and asking to touch her hand. Her response is left as a brief, unresolved stillness behind the panel.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved screens, round stone table with the unlit candle, luohan couch, hanging scroll, and single-source oil lamp lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the flowing white robe, pale complexion, long black hair, and delicate features cited from <Picture 2> are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark robe, black head wrap, black boots, \"捕\" waist tag, and dao sword cited from <Picture 3> are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber lamp light against deep shadow and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens across the center, the round stone table in the foreground carrying an unlit candle, a luohan couch along the right wall, and a hanging scroll above it. A single oil lamp burns on a low table, its amber flame the only source of light in the room. <Subject 2>, the constable in the dark robe, black head wrap, and dao sword, stands at the near side of the screens facing them, his weight planted on the stone floor. Through a narrow vertical gap between two central panels, the pale hand and white sleeve of <Subject 1> rest motionless at the edge of the amber light. The camera pushes in with small amplitude at slow speed toward that gap. <Subject 2> raises his right hand and turns it palm up, stopping a short distance short of the gap, his fingers slightly open and not advancing further. <Subject 2> (S1) says in a low, slightly tentative voice: <d>[Chinese] 让我摸摸手呗</d> He holds his open palm where it is, his gaze fixed on the gap. Behind the panel, the pale hand of <Subject 1> stills, her fingers curling faintly without withdrawing, while her silhouette behind the carved wood stays otherwise motionless. The lamp flame sways once and settles, leaving the two hands separated by the narrow gap of carved wood as the shot holds.\n\noverall_soundscape: Near-silence. The oil lamp wick gives a faint continuous sputter, a single quiet creak comes from the old wooden screen frame, and a subtle rustle of silk shifts once behind the panels.\n\nnon_diegetic_music: A sustained low erhu drone at a very slow tempo, with one sparse plucked note that fades without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787850191899,
-    "promptEn": "constable reaches out to touch her hand through the screen gap, no response",
-    "lastPlayedAt": 1788327330342,
-    "playCount": 21
-  },
-  {
-    "id": "9bd8c9fbf1f6",
-    "source": "h3console",
-    "prompt": "捕快跟屏风后的女子说，今晚愿意相信她",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with long black hair, a delicate face of unnaturally pale complexion, and a still, unhurried bearing.\n<Subject 2> is the constable in <Picture 3>, wearing a dark long robe with a dao sword at his waist, a black head cloth, black boots, resolute features, and a waist badge reading \"捕\".\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens across the center, a round stone table, a luohan bed along the right wall, and a hanging scroll on the plaster wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing in the room established by <Picture 1>, facing the carved screens where <Subject 1> stands behind, and telling her that he chooses to trust her tonight. <Picture 1> anchors the first frame and the room's composition; <Picture 2> and <Picture 3> define the two characters' appearances only.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved screens, round stone table, luohan bed, hanging scroll, and dim amber lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the white robe, long black hair, pale complexion, and still bearing are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark long robe, dao sword, black head cloth, and waist badge are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber lamp light against deep shadow and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. Five carved wooden screens stand across the center of the room, a luohan bed sits along the right wall, a hanging scroll hangs on the plaster wall, and an unlit candle stands on the round stone table. The only light comes from a single oil lamp on a low wooden table, its amber flame swaying gently and casting soft moving shadows across the carved screens. <Subject 2>, the constable in a dark long robe with a dao sword at his waist and a black head cloth, stands in the foreground facing the screens. The camera pushes in with small amplitude at slow speed toward the narrow gaps between the screen panels. Behind them, the pale silhouette of <Subject 1>, the woman in a flowing white robe with long black hair, stands motionless, her face half-lit by the amber glow. <Subject 2> (S1) steadies his breathing and eases his stance, his shoulders lowering slightly, and says in a low, quiet voice, <d>[Chinese] 好，今夜我先信你。</d> The words settle into the stillness of the room. Behind the screens, a faint rustle of silk crosses the gap and <Subject 1>'s white sleeve shifts once, her pale outline tilting slightly toward the sound. The oil lamp flame bends with the faint draft, sending a slow wave of shadow across the carved wood, then settles back to a gentle sway while <Subject 2> remains standing where he is, his gaze held on the screens.\n\noverall_soundscape: A faint continuous sputter from the oil lamp wick is the only steady sound in the room. A soft rustle of silk passes behind the screens once, and the old wooden floor creaks quietly under the constable's boots.\n\nnon_diegetic_music: A sustained low erhu drone at a very slow tempo, joined by a single sparse pipa pluck that fades without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787847447661,
-    "promptEn": "constable tells the woman behind the screen he'll trust her tonight",
-    "lastPlayedAt": 1788328644960,
-    "playCount": 21
-  },
-  {
-    "id": "8cb493338cb3",
-    "source": "h3console",
-    "prompt": "捕快答应屏风后的女子，只要信他就送她安全出去",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens standing across its center, a round stone table holding an unlit candle, and deep shadow along the plaster walls.\n<Subject 1> is the mysterious woman in flowing white ancient robes with a delicate, unnaturally pale face and long black hair, standing behind the carved screens, as established in <Picture 2>.\n<Subject 2> is the male constable in a dark long robe with a dao sword at his waist, a black head wrap, dark boots, and a square waist placard bearing the character \"捕\", with stern resolute features, as established in <Picture 3>.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing before the five carved screens of <Picture 1>, quietly promising <Subject 1>, still hidden behind the panels, that he will see her safely out of the room if she will trust him. The oil lamp remains the room's only light source while the camera slowly pushes in between the two figures.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, the five carved screens, the stone table, and the warm amber single-source lamp lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the flowing white robe, pale complexion, long black hair, and still bearing behind the screens are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark robe, dao sword, black head wrap, waist placard, and stern features are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with a warm amber oil lamp as the room's only light source against deep surrounding shadow.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide view of the dim old room. Five carved wooden screens stand across the center of the space, a round stone table sits beside them with an unlit candle resting on its surface, and the single oil lamp throws warm amber light across the floorboards while the far corners sink into darkness. <Subject 2>, the constable in a dark robe with a dao sword at his waist and a square waist placard bearing the character \"捕\", stands a few paces from the table facing the screens, his weight planted evenly on the boards. Behind the carved panels, <Subject 1> waits in shadow, only the pale curve of her cheek and the edge of one white sleeve visible through the narrow vertical gaps. His chest rises once with a steadying breath, and his right hand drops away from the sword hilt as he decides not to reach for it. The camera begins a slow push in with small amplitude at slow speed toward the screens as <Subject 2> lowers his voice to a steady, quiet register intended for her alone. <Subject 2> (S1) says: <d>[Chinese]只要你肯信我，我会护着你走出去。</d>\nHe lets the words hang there, gaze fixed on the gap where her face is half-lit by the amber glow. Beyond the carved lattice, <Subject 1>'s silhouette holds still for a breath, then the white sleeve drifts once behind the panel, neither drawing closer nor retreating. The oil lamp flame sways gently, sending a slow wave of light and shadow across the carved screen surface while the candle on the stone table stays dark. The camera continues its slow forward drift until the frame settles on the carved pattern that separates the two figures.\n\noverall_soundscape: A faint sputter from the oil lamp wick continues beneath a hushed indoor room tone. Night wind presses softly against the old window lattice, a single board creaks underfoot early on, and a brief rustle of silk comes from behind the carved screens.\n\nnon_diegetic_music: A sustained low erhu drone at a very slow tempo, joined by one sparse plucked string tone that fades away without resolving.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787831362083,
-    "promptEn": "constable promises to get her out safely if she trusts him",
-    "lastPlayedAt": 1788328692996,
-    "playCount": 21
-  },
-  {
-    "id": "5ede61c01b22",
-    "source": "h3console",
-    "prompt": "女子在屏风后让捕快出去看看，他握刀警惕地走出去",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, with a flowing white traditional Chinese robe, long black hair, and an unnaturally pale, delicate face.\n<Subject 2> is the young male constable in <Picture 3>, with a dark robe, a \"捕\" waist plate, a dao sword at his waist, black boots, and a black headwrap, with a resolute face.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens in the center, a single oil lamp on a round stone table, an arhat bed against the right wall, and hanging scrolls on the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing before the carved screens in the room established by <Picture 1>. <Subject 1>, hidden behind the screens, speaks in a low off-screen voice and orders him to go outside and look, and <Subject 2> turns toward the wooden door, one hand on his sword, and steps outside with wary caution.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, oil lamp, arhat bed, and hanging scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the flowing white robe, long black hair, and unnaturally pale complexion are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark robe, \"捕\" waist plate, dao sword, black boots, and black headwrap are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style, with warm amber oil-lamp light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens standing across the center, a single oil lamp casting flickering amber light across the round stone table, an arhat bed against the right wall, and hanging scrolls along the walls. <Subject 2>, the young constable in a dark robe with a \"捕\" waist plate and a dao sword at his waist, stands in the midground with his back to the camera, facing the carved screens. The camera holds a static shot. Through the narrow gaps between the screen panels, <Subject 1>, the pale woman in a flowing white robe with long black hair, is faintly visible, her face half-lit by the amber glow. <Subject 1> (S1) says in an off-screen voiceover from behind the carved screens, in a low, calm voice with an eerie stillness: <d>[Chinese] 你出去看看。</d> while her lips remain closed. The oil lamp flame bends, sending a slow wave of shadow across the screen surface. <Subject 2> tenses, his shoulders rising beneath the dark robe, and slowly turns his head toward the heavy wooden door at the far edge of the room, one hand drifting to the dao hilt at his waist.\n[Shot 2] At 00:04.000, the camera cuts to a side angle of <Subject 2>, the constable in the dark robe from Shot 1, now turned toward the heavy wooden door. The camera trucks right with small amplitude at slow speed as he steps forward deliberately, his right hand steady on the dao hilt, his black boots crossing the stone floor. Behind him, the oil lamp throws his long shadow across the carved screens. He pauses at the threshold, head half-turned back toward the room, then pulls the door open and steps outside. As the door swings, the pale silhouette of <Subject 1> remains motionless between the screen panels, and the shot holds on the empty room as the flame steadies.\n\noverall_soundscape:\nBoot steps cross the stone floor with a deliberate, measured rhythm, followed by the heavy creak of the wooden door opening. The oil lamp wick sputters faintly. A single rustle of silk comes from behind the carved screens as the woman shifts, then silence.\n\nnon_diegetic_music:\nA sustained low erhu drone at a slow tempo, joined by a single sparse pipa pluck that fades without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787762342686,
-    "promptEn": "woman tells him from behind the screen to go check outside, he steps out warily",
-    "lastPlayedAt": 1788331144924,
-    "playCount": 22
-  },
-  {
-    "id": "47a6de809c61",
-    "source": "h3console",
-    "prompt": "屏风后女子动了动，转身轻声安慰捕快",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious pale woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with long black hair, delicate features, and an unnaturally pale complexion.\n<Subject 2> is the young male constable in <Picture 3>, wearing a dark robe with a dao sword at his waist, a \"捕\" waist plaque, black boots, and a black headscarf.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens, a single burning candle on a round stone table, an arhat bed on the right side, and a hanging scroll on the far wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing before the carved screens in the room established by <Picture 1>. Behind the screens, <Subject 1> stirs, turns toward him, and softly speaks a line of reassurance, her pale face visible through the narrow gaps between the panels while the candle flame flickers across the dim room.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, candle-lit round stone table, arhat bed, hanging scroll, and dim amber lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the flowing white robe, long black hair, delicate features, and pale complexion are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark robe, dao sword, \"捕\" waist plaque, black boots, and black headscarf are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber candlelight against deep surrounding shadow and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens in the midground, a single candle burning on the round stone table, an arhat bed against the right wall, and a hanging scroll on the far wall. <Subject 2>, the constable in the dark robe with a dao sword at his waist, a \"捕\" waist plaque, black boots, and a black headscarf, stands in the foreground facing the screens, his weight settled and his right hand resting on the sword hilt. Behind the panels, the white silhouette of <Subject 1> stirs and turns toward him, a slow shift of cloth in the gloom. The candle flame wavers, throwing long shadows across the carved screen surfaces, and the camera pushes in with small amplitude at slow speed toward the screens.\n[Shot 2] At 00:03.500, the camera cuts to a close-up of <Subject 1>, the pale woman in the flowing white robe, seen through a narrow gap between two screen panels. The carved wood frames her face on both sides, candlelight catching her long black hair and the unnaturally pale skin of her cheek. <Subject 1> (S1) says in a soft, gentle voice at a calm, slow pace: <d>[Chinese] 有你在，我安心些了。</d> As she finishes, a faint, almost imperceptible smile settles on her lips while a single strand of black hair shifts across her cheekbone. The candle flame bends once, briefly deepening the shadow in her eye socket before returning to stillness, and the shot holds on her face.\n\noverall_soundscape: The candle flame emits a faint continuous crackle beneath the soft rustle of the woman's silk robe and the constable's slow, controlled breathing. Distant wind presses against the exterior walls, and old wood creaks once.\n\nnon_diegetic_music: A single sustained low dizi drone at a very slow tempo, joined by a sparse guzheng pluck that fades without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787762332289,
-    "promptEn": "woman behind the screen stirs and softly reassures him",
-    "lastPlayedAt": 1788330492387,
-    "playCount": 22
-  },
-  {
-    "id": "0abf6e529c92",
-    "source": "h3console",
-    "prompt": "捕快坐在小凳上端着水杯，慢慢抬眼看向屏风后的女子",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious white-robed woman in <Picture 2>, with long black hair, unnaturally pale skin, delicate features, and a flowing white traditional Chinese robe.\n<Subject 2> is the young constable in <Picture 3>, with a dark long robe, a dao sword at his waist, a black headcloth, black boots, and a carved waist token reading \"捕\".\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens in the center, a single oil lamp on the round stone table, a couch bed on the right, and hanging wall scrolls in shadow.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> sitting quietly on a low wooden stool beside the round stone table, holding a ceramic cup of water in both hands, then slowly raising his gaze toward <Subject 1>, who stands motionless behind the carved wooden screens. The oil lamp flame sways gently between them as the silence holds.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the dim room layout, five carved screens, round stone table, oil lamp, couch bed, and wall scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, unnaturally pale skin, and delicate features are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark long robe, dao sword, black headcloth, black boots, and waist token are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp light against deep surrounding darkness and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium shot of the dim old room. Five carved wooden screens stand in the midground, a single oil lamp on the round stone table casts flickering amber light, and a couch bed and hanging scrolls recede into shadow along the right wall. <Subject 2>, the young constable in a dark long robe with a dao sword at his waist and a black headcloth, sits on a low wooden stool beside the stone table, holding a plain ceramic cup of water in both hands. His back is straight, his shoulders still, his gaze lowered toward the cup's surface where the lamp light reflects faintly. Behind the screens, the faint pale silhouette of <Subject 1>, the white-robed woman with long black hair, stands motionless in the amber-lit darkness. The camera pushes in with small amplitude at slow speed, the silence holding as the water trembles slightly from the stillness of his hands. [Shot 2] At 00:04.000, the camera cuts to a close-up of <Subject 2>'s face from a slightly low angle. He slowly raises his eyes, lifting his gaze from the water toward the carved screens, his brow level and his jaw firm. The amber lamp light catches the line of his cheekbone while deep shadow fills the rest of his face. Through the narrow gaps between the screen panels, <Subject 1>, the white-robed woman with long black hair and unnaturally pale skin, is visible, standing perfectly motionless with her pale face half-lit by the wavering flame. The camera holds a static shot as his gaze meets hers across the dim room, and the oil lamp flame bends once, sending a slow wave of shadow across the carved wood.\n\noverall_soundscape:\nA faint continuous sputter from the oil lamp wick over low indoor room tone. A soft rustle of dark fabric accompanies the constable's slow movement as he raises his head, and the aged wooden floor creaks once beneath him.\n\nnon_diegetic_music:\nA single sustained low string drone at a very slow tempo, joined by one sparse guzheng harmonic that swells faintly as the constable lifts his gaze, then fades without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787762311804,
-    "promptEn": "constable sits with a cup of water, slowly looking up at the woman behind the screen",
-    "lastPlayedAt": 1788331439169,
-    "playCount": 22
-  },
-  {
-    "id": "850f13881576",
-    "source": "h3console",
-    "prompt": "捕快隔着屏风盯着女子，抿着嘴不说话",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with long black hair, pale skin, and a refined face that carries an unnatural pallor.\n<Subject 2> is the young male constable in <Picture 3>, with a dark indigo robe, a black headscarf, black boots, a dao sword at his waist, a waist tag reading \"捕\", and a resolute, stern face.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens, a single oil lamp on a round stone table, a daybed against the right wall, and deep shadows across the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video captures a tense silence as <Subject 2> holds <Subject 1>'s gaze through the carved screens without speaking. He slightly purses his lips and keeps his eyes fixed on her, offering no answer. The camera slowly pushes in on his face while the oil lamp flame wavers, then cuts to <Subject 1>'s eyes watching him through the screen gaps. Neither figure moves or speaks.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, oil lamp, daybed, and wall-scroll ambience are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, pale complexion, and unnaturally still bearing are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark indigo robe, black headscarf, dao sword, waist tag, and resolute expression are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with a warm single-source oil lamp against deep surrounding darkness and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium shot of the dim old room. <Subject 2>, the constable in a dark indigo robe with a dao sword at his waist and a waist tag reading \"捕\" glinting at his belt, stands facing the carved wooden screens with his weight evenly planted on the stone floor. Through the narrow vertical gaps between screen panels, <Subject 1>, the woman in a white robe with long black hair, is partially visible, her pale face half-lit by the amber oil lamp glow. The camera pushes in with small amplitude at slow speed toward <Subject 2>'s face. His jaw tightens and his lips press together into a thin, slight purse. He does not speak. His eyes remain fixed on the gaps where <Subject 1>'s face is visible, her gaze steady and waiting on him. The oil lamp flame bends slightly, sending a slow wave of shadow across the carved screen surface. He continues to look at her with lips pressed, giving no answer. [Shot 2] At 00:03.500, the camera cuts to an extreme close-up of <Subject 1>'s eyes seen through a narrow gap between two screen panels. The carved wood frames her gaze on both sides. The amber lamp light catches the surface of her dark iris. Her expression is perfectly neutral but alert, waiting for his answer. A single strand of black hair shifts across her cheekbone from an imperceptible air current. The camera holds a static shot. The oil lamp light pulses once, briefly deepening the shadow in her eye socket before returning.\n\noverall_soundscape: Near-silence. The oil lamp wick emits a faint continuous sputter. A single quiet creak comes from the old wooden screen frame. Distant wind presses against the exterior wall.\n\nnon_diegetic_music: N/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787762286261,
-    "promptEn": "constable stares at her through the screen, silent, saying nothing",
-    "lastPlayedAt": 1788329543604,
-    "playCount": 21
-  },
-  {
-    "id": "59711a102e84",
-    "source": "h3console",
-    "prompt": "捕快站在油灯下，隔着屏风跟女子对视不退",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, with unnaturally pale skin, long black hair, and a flowing white traditional Chinese robe, standing motionless behind the carved wooden screens.\n<Subject 2> is the young male constable in <Picture 3>, with a resolute face, wearing a dark long robe, a dao sword at his waist, black boots, and a black headwrap.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens in the center, a single oil lamp on a round stone table, a luohan bed against the right wall, and hanging scrolls on the far wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing frozen beneath the oil lamp in the room established by <Picture 1>, holding <Subject 1>'s gaze through the gaps of the carved screens without stepping back. The two figures remain locked in stillness as the lamp flame wavers, casting long shifting shadows across the carved wood.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved screens, oil lamp, round stone table, luohan bed, and hanging scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the woman's white robe, unnaturally pale skin, long black hair, and motionless posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the constable's dark robe, dao sword, black boots, black headwrap, and resolute bearing are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber lamp light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. Five carved wooden screens stand in the midground; a single oil lamp on the round stone table throws flickering amber light across the floor; a luohan bed rests against the right wall and hanging scrolls line the far wall. <Subject 2>, the constable in <Picture 3>, stands beneath the lamp's glow, his dark robe and dao sword at his waist plainly visible, his body rigid and his weight planted. His gaze is fixed on the narrow gap between two screen panels where <Subject 1>, the woman in the white robe with unnaturally pale skin, is partially visible. The camera holds a static shot with small amplitude shake. The lamp flame wavers once, sending long shadows rippling across the carved screen surface, but <Subject 2> does not move and does not step back; only the slow rise and fall of his chest disturbs the stillness.\n[Shot 2] At 00:03.500, the camera cuts to a close-up of <Subject 2>'s face. The warm lamp light catches the hard line of his jaw while deep shadow fills the rest of the frame. His eyes remain locked on <Subject 1> behind the screens, and a faint pale reflection of her white robe slides across the dark surface of his iris. His right hand rests on the dao hilt at his waist without drawing. The camera pushes in with small amplitude at slow speed as he holds her gaze, unmoving, refusing to retreat. The silence holds through the final frame.\n\noverall_soundscape:\nNear-silence broken by the faint continuous sputter of the oil lamp wick. The old wooden screen frame creaks once. A barely audible rustle of silk comes from behind the screens, and distant wind presses against the exterior wall.\n\nnon_diegetic_music:\nA sustained low erhu drone at a slow tempo with a single sparse high note that hangs unresolved, gradually increasing in volume and holding until the final frame.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787762224218,
-    "promptEn": "constable stands under the lamp, holding her gaze through the screen, not backing away",
-    "lastPlayedAt": 1788329579630,
-    "playCount": 21
-  },
-  {
-    "id": "f2e70205766b",
-    "source": "h3console",
-    "prompt": "捕快小心检查门窗有没有锁好，女子在屏风后看着他",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with pale skin, long black hair, and an unnaturally still posture.\n<Subject 2> is a young male constable in <Picture 3>, with a resolute face, wearing a dark indigo robe with a dao sword at his waist, a black belt plaque engraved with \"捕\", black boots, and a black headband.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens, a single oil lamp on a low wooden table, a round stone table bearing a cold unlit candle, a luohan bed on the right, and hanging scrolls on the cracked walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> moving cautiously across the room established by <Picture 1> to inspect the wooden door and the paper-lattice window, testing their fastenings while staying alert. <Subject 1> watches him from behind the carved screens as the oil lamp light shifts.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, round stone table, luohan bed, hanging scrolls, and oil lamp lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the flowing white robe, long black hair, pale complexion, and still posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark indigo robe, black belt plaque engraved with \"捕\", dao sword, black boots, and black headband are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber key light from a single oil lamp against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. Five carved wooden screens stand in the midground, their dark frames layered over the plaster walls; near the center, a low wooden table holds the single oil lamp, its flame casting flickering amber light across the round stone table bearing a cold unlit candle, the luohan bed on the right, and the hanging scrolls on the cracked walls. <Subject 2>, the young constable in the dark indigo robe with a dao sword at his waist, a black belt plaque engraved with \"捕\", and a black headband, stands in the foreground with his weight low and his right hand resting near the sword hilt. His resolute face turns slowly across the room as his eyes scan the far wall. He steps forward toward the heavy wooden door on the left, his black boots scuffing softly on the stone floor. The camera tracks him with small amplitude at slow speed as he reaches the door, runs his open palm over the rough planks, and tests the iron latch with two fingers, his knuckles flexing. Through the narrow gaps between the screen panels, <Subject 1>, the pale woman in the flowing white robe with long black hair, stands perfectly motionless, her face half-lit by the amber glow. The oil lamp flame wavers and throws long, shifting shadows across the screens as <Subject 2> leans in and presses his ear close to the plank seams, listening.\n[Shot 2] At 00:04.500, the camera cuts to a close-up of <Subject 2>'s right hand gripping the iron latch bolt, knuckles whitening as he tests it once while the wood creaks under the pressure. The camera pushes in with small amplitude at slow speed. He releases the latch and turns his head sharply toward the paper-lattice window on the right wall. Faint cold moonlight bleaches the window paper from outside while a slow shadow slides across its surface from beyond the pane. He steps to the window, lifts one finger, and presses the lattice frame, checking for gaps between the paper and the wood. Behind him, through the screen gaps, <Subject 1> has turned her head to follow him, still and silent, her pale face barely catching the edge of the lamp light.\n\noverall_soundscape:\nSlow, cautious boot steps on stone with a faint leather creak from the sword belt. The wooden door planks creak under the pressure of his palm, and the iron latch clinks once. The oil lamp wick sputters, and a subtle rustle of silk comes from behind the carved screens.\n\nnon_diegetic_music:\nA sustained low erhu drone at a slow tempo with sparse pipa plucks, growing faintly in volume as he tests the latch and fading out without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787738044140,
-    "promptEn": "constable carefully checks the door and window latches, she watches from behind the screen",
-    "lastPlayedAt": 1788329309405,
-    "playCount": 21
-  },
-  {
-    "id": "cfbef6e37ae4",
-    "source": "h3console",
-    "prompt": "捕快抓起油灯砸向墙上的美人画，画卷着火了",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with delicate but unnaturally pale features, long black hair, and an unsettling stillness.\n<Subject 2> is the young male constable in <Picture 3>, wearing a dark long robe, with a waist token reading \"捕\", a dao sword at his belt, black boots, and a black headwrap, and a resolute, stern face.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens, a round stone-top table holding a single lit oil lamp, an arhat bed along the right wall, and a hanging beauty portrait on the background wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> silently seizing the oil lamp from the table in the room established by <Picture 1> and hurling it at the hanging beauty portrait. The lamp strikes the portrait, oil splashes across the painted figure, and the scroll catches fire, the new flame-light flaring across the dim room and illuminating <Subject 1> standing motionless behind the carved screens.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, stone-top table, arhat bed, and hanging scroll are retained as the spatial anchor for the opening composition.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, pale features, long black hair, and motionless posture are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark robe, waist token, dao sword, black boots, and black headwrap are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with a single warm amber oil-lamp key light against deep surrounding shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. Five carved wooden screens stand in the midground, their lacquered frames catching faint amber glints, while a round stone-top table holding a single lit oil lamp sits near the center and an arhat bed runs along the right wall. A hanging beauty portrait hangs on the background wall, half-lost in shadow. <Subject 2>, the young male constable in a dark long robe with a waist token reading \"捕\", a dao sword at his belt, black boots, and a black headwrap, stands beside the table, rigid and silent, his jaw set and his eyes fixed on the portrait. Behind the carved screens, the pale white figure of <Subject 1>, the woman in a flowing white robe with long black hair, is faintly visible in the gaps between the panels, utterly still. <Subject 2> remains silent. He reaches down, grips the oil lamp by its metal base, and lifts it from the table, the flame swaying and throwing long shadows across the walls as the camera pushes in with small amplitude at slow speed. He draws his arm back across his body, hesitates for half a beat, and hurls the lamp toward the hanging beauty portrait.\n[Shot 2] At 00:04.000, the camera cuts to the oil lamp striking the hanging beauty portrait. The lamp shatters against the scroll, ceramic shards and oil scattering across the wall, and flames catch the painted silk, flaring upward with a sudden orange glow. The camera pushes in with small amplitude at slow speed as the firelight sweeps across the room, throwing long, dancing shadows of the carved screens across the walls. A pool of burning oil spreads along the floor as the painted figure blackens and curls at the edges. The climbing flames throw narrow stripes of orange light across <Subject 1>'s white robe as she stands motionless in the gap between the screens, her pale face lit by the fire.\n\noverall_soundscape:\nLow room tone of an old wooden house settles beneath soft creaks from the floorboards as the constable shifts his weight. The metal base of the oil lamp scrapes across the stone tabletop, then the lamp cuts through the air with a low whoosh. A sharp ceramic shatter and a wet oil splatter meet the wall, followed by the rising crackle and low hiss of flames catching the painted scroll.\n\nnon_diegetic_music:\nA sustained low guzheng drone at a slow tempo with sparse, high pipa plucks, holding an uneasy stillness. As the lamp strikes the portrait, a sharp percussion hit lands and a fast tremolo erhu line flares upward before cutting off abruptly.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787730413689,
-    "promptEn": "constable grabs the oil lamp and hurls it at the portrait, the scroll catches fire",
-    "lastPlayedAt": 1788329005229,
-    "playCount": 21
-  },
-  {
-    "id": "e6c23c6adc42",
-    "source": "h3console",
-    "prompt": "捕快雨夜走进屋，发现屏风后的女子，编了个身份介绍自己",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with a pale, delicate face, long black hair, and an unnaturally still posture.\n<Subject 2> is the young constable in <Picture 3>, with a dark indigo robe, a dao sword at his waist, a black headscarf, and a resolute jaw.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens, a low round stone table holding a single lit oil lamp, and deep shadows across the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> stepping in from the rainy night into the dim room established by <Picture 1>. He scans the shadows, notices <Subject 1> standing motionless behind the carved screens, and introduces himself with a guarded cover story. The oil lamp flame wavers as he speaks, and <Subject 1> remains perfectly still.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, round stone table, oil lamp, and deep-shadow lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the white robe, long black hair, pale face, and motionless posture are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark indigo robe, dao sword, black headscarf, and resolute features are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens standing in the midground and a single oil lamp casting flickering amber light across a low round stone table. The camera pushes in with small amplitude at slow speed as the wooden door at the right edge of frame creaks open. <Subject 2>, the young constable in a dark indigo robe with a dao sword at his waist and a black headscarf, steps inside, his robe and hair damp from the night rain, droplets sliding from the edge of his headscarf. His boots leave damp prints on the stone floor. He closes the door behind him, muffling the storm, and stands still for a beat, his gaze sweeping the shadowed room before settling on the carved screens. Through the narrow gaps between the screen panels, the pale silhouette of <Subject 1>, the woman in a flowing white robe with long black hair, stands perfectly motionless, half-lit by the amber glow. The constable's jaw tightens and his right hand drifts toward the dao hilt without drawing. <Subject 2> (S1) steadies his breathing and says in a guarded, low, tired voice, <d>[Chinese] 我姓陈，只是个赶路的书生，雨夜里不知被什么东西缠上了。</d> As his words settle, the oil lamp flame bends, sending a slow wave of shadow across the carved screens, while <Subject 1> remains perfectly still behind the wood.\n\noverall_soundscape:\nRain pounds against the roof and walls before the door closes, then settles into a muffled drumming from outside. The door hinge creaks, damp robe fabric rustles with the constable's cautious steps on the stone floor, and the oil lamp wick sputters softly.\n\nnon_diegetic_music:\nA sustained low erhu drone at a slow tempo with a sparse pipa pluck, the drone gradually gaining tension and fading to silence as the constable finishes speaking.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787730331754,
-    "promptEn": "constable steps in from the rainy night, spots her, introduces himself with a cover story",
-    "lastPlayedAt": 1788326397551,
-    "playCount": 21
-  },
-  {
-    "id": "d720d52c783f",
-    "source": "h3console",
-    "prompt": "捕快雨夜走进屋，发现屏风后的女子，编了个身份介绍自己",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with a pale, delicate face, long black hair, and an unnaturally still posture.\n<Subject 2> is the young constable in <Picture 3>, with a dark indigo robe, a dao sword at his waist, a black headscarf, and a resolute jaw.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens, a low round stone table holding a single lit oil lamp, and deep shadows across the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> stepping in from the rainy night into the dim room established by <Picture 1>. He scans the shadows, notices <Subject 1> standing motionless behind the carved screens, and introduces himself with a guarded cover story. The oil lamp flame wavers as he speaks, and <Subject 1> remains perfectly still.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, round stone table, oil lamp, and deep-shadow lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the white robe, long black hair, pale face, and motionless posture are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark indigo robe, dao sword, black headscarf, and resolute features are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens standing in the midground and a single oil lamp casting flickering amber light across a low round stone table. The camera pushes in with small amplitude at slow speed as the wooden door at the right edge of frame creaks open. <Subject 2>, the young constable in a dark indigo robe with a dao sword at his waist and a black headscarf, steps inside, his robe and hair damp from the night rain, droplets sliding from the edge of his headscarf. His boots leave damp prints on the stone floor. He closes the door behind him, muffling the storm, and stands still for a beat, his gaze sweeping the shadowed room before settling on the carved screens. Through the narrow gaps between the screen panels, the pale silhouette of <Subject 1>, the woman in a flowing white robe with long black hair, stands perfectly motionless, half-lit by the amber glow. The constable's jaw tightens and his right hand drifts toward the dao hilt without drawing. <Subject 2> (S1) steadies his breathing and says in a guarded, low, tired voice, <d>[Chinese] 我姓陈，只是个赶路的书生，雨夜里不知被什么东西缠上了。</d> As his words settle, the oil lamp flame bends, sending a slow wave of shadow across the carved screens, while <Subject 1> remains perfectly still behind the wood.\n\noverall_soundscape:\nRain pounds against the roof and walls before the door closes, then settles into a muffled drumming from outside. The door hinge creaks, damp robe fabric rustles with the constable's cautious steps on the stone floor, and the oil lamp wick sputters softly.\n\nnon_diegetic_music:\nA sustained low erhu drone at a slow tempo with a sparse pipa pluck, the drone gradually gaining tension and fading to silence as the constable finishes speaking.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787730331663,
-    "promptEn": "constable steps in from the rainy night, spots her, introduces himself with a cover story",
-    "lastPlayedAt": 1788327852741,
-    "playCount": 21
-  },
-  {
-    "id": "84e8337de242",
-    "source": "h3console",
-    "prompt": "第一人称视角冲进屋，回头害怕地看向门口，女子被吓了一跳",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the single pale-skinned woman in <Picture 2>, wearing a sheer white flowing traditional robe, with a high dark hair bun, delicate features, and an unnaturally still, watchful bearing.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens across the center, a single oil lamp on the round stone table, a wooden daybed along the right wall, and hanging scrolls on the shadowed plaster walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows the player bursting into the room from a first-person point of view and glancing back toward the doorway in fear, while <Subject 1> reacts with a single startled expression at the sudden intrusion.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, round stone table, oil lamp, daybed, wall hangings, and amber shadow lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the sheer white robe, dark hair bun, pale complexion, and still bearing are retained as a single figure; only her expression changes to visible shock.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber lamp light against deep surrounding darkness and a slightly desaturated color palette.\n[Shot 1] The scene begins from the composition established by <Picture 1>, a first-person point-of-view shot of the dim old room: five carved wooden screens stand across the center, a single oil lamp burns on the round stone table, a wooden daybed lines the right wall, and hanging scrolls catch faint amber light on the shadowed plaster walls. The camera, representing the player's own viewpoint, sways unsteadily forward as if just having burst through the door, then swings back toward the entrance behind it with a quick, panicked motion. Breathing hard, the player says in a shaky, urgent voice, <d>[Chinese] 对不起，我不是有意闯进来的，外面有东西在追我。</d> The camera turns back into the room to reveal <Subject 1>, the single pale woman in the sheer white robe standing beyond the carved screens. Her eyes widen and her lips part slightly in visible surprise at the sudden intrusion, one hand rising partway toward her chest, her otherwise still and composed bearing giving way to that one startled reaction. The oil lamp flame flickers from the sudden draft through the open door.\n\noverall_soundscape:\nUrgent footsteps and ragged breathing accompany a brief creak from the door swinging on its hinges. The oil lamp wick sputters faintly, joined by a quiet groan from the old wooden floorboards.\n\nnon_diegetic_music:\nA sudden low string stinger at the moment the player bursts in, settling into a slow, tense sustained drone with widely spaced pipa plucks.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787724313620,
-    "promptEn": "first-person view bursting into the room, glancing back in fear, the woman startled",
-    "lastPlayedAt": 1788326261435,
-    "playCount": 21
-  },
-  {
-    "id": "81b6c3789111",
-    "source": "h3console",
-    "prompt": "捕快不再四处张望，转身面对屏风，跟女子的影子说话",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious pale-faced woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with long black hair and an unnaturally still posture.\n<Subject 2> is the young constable in <Picture 3>, with a resolute, angular face, a dark indigo robe, a black headscarf, a dao sword and a waist badge reading \"捕\" at his belt, and black boots.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens, a single oil lamp on a round stone table, a low wooden bed on the right, and wall scrolls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> stopping his cautious scan of the dim room, turning to face the carved screens, and speaking to <Subject 1>'s silhouette as he resolves not to look elsewhere. He holds her still figure in view as the oil lamp flame continues to waver.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, oil lamp, low wooden bed, wall scrolls, and dim lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, pale complexion, and motionless bearing behind the screens are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark indigo robe, black headscarf, dao sword, waist badge, and resolute features are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber lamplight against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens in the center, a single oil lamp burning on a round stone table, and wall scrolls hanging on the right. <Subject 2>, the constable in the dark indigo robe with a black headscarf and a dao sword at his waist, stands near the center of the room with his head turned toward the wall scrolls, finishing a wary scan. He holds the posture for a beat, then squares his shoulders, turns deliberately to face the screens, and lowers his gaze. Behind the gaps between the screen panels, the faint silhouette of <Subject 1>, the woman in a white robe with long black hair, stands perfectly motionless. <Subject 2> (S1) says in a low, steady voice with a resolute cadence, <d>[Chinese] 好，我先不看别处了。</d> He closes his lips and keeps his eyes fixed on the screens, his right hand resting on the dao hilt.\n[Shot 2] At 00:04.000, the camera cuts to a close-up of <Subject 2>'s face from a slightly low angle. The oil lamp casts flickering amber light across his jaw and cheekbone while his eyes remain locked on the screens, where <Subject 1>'s pale face is partially visible through a narrow gap between panels. The camera pushes in with small amplitude at slow speed. His brow is set, his lips pressed into a firm line. The oil lamp flame bends once, sending a slow wave of shadow across his face, and he does not look away.\n\noverall_soundscape:\nThe oil lamp wick emits a faint, continuous sputter while low wooden creaks sound from the old room. The constable's robe rustles softly as he turns, and his boots shift once on the stone floor. A distant draft presses against the exterior wall.\n\nnon_diegetic_music:\nA sustained low erhu drone at a slow tempo, with a single sparse guzheng pluck that decays without resolution as the constable holds his gaze.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787708713767,
-    "promptEn": "constable stops looking around, turns to face the screen, speaks to her silhouette",
-    "lastPlayedAt": 1788326169354,
-    "playCount": 21
-  },
-  {
-    "id": "5dc95a15cd5d",
-    "source": "h3console",
-    "prompt": "捕快轻声跟屏风后的女子说，愿意留下来陪她",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the woman in <Picture 2>, a mysterious woman in a flowing white traditional Chinese robe, with unnaturally pale skin, long black hair, and a delicate yet eerily still face.\n<Subject 2> is the man in <Picture 3>, a young constable in a dark robe with a black headscarf, black cloth boots, a belt plaque reading \"捕\", a dao sword at his waist, and a resolute, composed face.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens at the center, a single oil lamp on a round stone table holding a candle, an arhat bed on the right side, and hanging wall scrolls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing before the carved screens and speaking in a gentle, steady voice to <Subject 1>, who remains unnaturally still behind them. He keeps his hands open at his sides and offers to stay and keep her company, while the dim oil-lamp-lit room established by <Picture 1> frames the exchange.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the dim interior, five carved screens, oil lamp, round stone table, arhat bed, and hanging wall scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the flowing white robe, unnaturally pale skin, long black hair, and still bearing are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark robe, black headscarf, belt plaque, dao sword, and resolute features are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room: five carved wooden screens stand at the center, a single oil lamp casts flickering amber light across the round stone table and the arhat bed on the right, and hanging wall scrolls fade into shadow. <Subject 2>, the constable in a dark robe with a black headscarf, a belt plaque reading \"捕\", and a dao sword at his waist, stands at the left edge of the frame facing the screens. Through the narrow gaps between the screen panels, <Subject 1>, the white-robed woman with unnaturally pale skin and long black hair, is partially visible, her face half-lit by the amber glow. The camera pushes in with small amplitude at slow speed. <Subject 2> (S1) takes one gentle step forward, keeps both hands open and visible at his sides, and speaks in a warm, steady voice with a calm, resolute delivery, <d>[Chinese] 若姑娘害怕，我陪你说会儿话。</d> As the words settle, the oil lamp flame bends slightly and sends a slow wave of shadow across the screen surface; <Subject 1> remains motionless behind the carved wood.\n[Shot 2] At 00:05.000, the camera cuts to an extreme close-up of <Subject 1>'s eyes framed by the narrow gap between two screen panels. The amber lamp light catches her dark iris through the carved wood, and her gaze is empty and unnervingly still. A single strand of black hair shifts across her cheekbone in the faint draft. The camera holds a static shot as the silence stretches; the oil lamp light pulses once, deepening the shadow in her eye socket before settling.\n\noverall_soundscape: The oil lamp wick emits a faint continuous sputter, and old wooden floorboards creak softly under the constable's single step. A low breeze presses against the exterior wall while the room settles into near-silence.\n\nnon_diegetic_music: A sustained low erhu drone at a slow tempo, joined by a single sparse pipa pluck after the constable's line, fading without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787708701547,
-    "promptEn": "constable gently tells her he'll stay and keep her company",
-    "lastPlayedAt": 1788331054852,
-    "playCount": 22
-  },
-  {
-    "id": "8d9c1dad041f",
-    "source": "h3console",
-    "prompt": "捕快应了邀请，挪到屏风旁边的小凳坐下",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with pale skin, long black hair, and an unnaturally still posture.\n<Subject 2> is the young male constable in <Picture 3>, with a stern face, wearing a dark robe, a waist tablet inscribed with the character \"捕\", a dao sword at his belt, black boots, and a black headscarf.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens, a round stone table with a single oil lamp, an arhat bed along the right wall, and deep shadows along the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> acknowledging an invitation and drawing closer to the carved screens: he says he will sit nearer, steps forward past the round stone table, and seats himself on a low wooden stool close to the screens, where <Subject 1> waits motionless behind the carved wood. The oil lamp flame wavers as he settles.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, round stone table, arhat bed, and wall shadows are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, pale complexion, and motionless posture are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark robe, \"捕\" waist tablet, dao sword, black boots, and black headscarf are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room: five carved wooden screens stand in the midground, a single oil lamp casts a small pool of amber light across the round stone table, and deep shadow pools along the walls. <Subject 2>, the stern-faced constable in the dark robe with a \"捕\" waist tablet and a dao sword at his belt, stands in the foreground facing the screens, his hands resting on his sword belt. The camera pushes in with small amplitude at slow speed as he speaks in a low, even young male voice (S1): <d>[Chinese] 好，我坐近些。</d> He nods once and steps forward past the round stone table. As he moves, the white silhouette of <Subject 1>, the woman in the flowing robe with long black hair, comes into view through the gaps between the screen panels, standing perfectly still.\n[Shot 2] At 00:03.500, the camera cuts to a medium shot as <Subject 2> lowers himself onto a low wooden stool drawn close to the screens, settling with a soft exhalation and bracing one hand on his thigh. The oil lamp light catches the line of his jaw while the rest of the room falls into shadow. He lifts his gaze to the screen gaps, where <Subject 1> remains motionless, her white robe faintly lit by the amber glow and her face half-turned toward him. The camera holds a static shot as the oil lamp flame wavers once, sending a slow wave of shadow across the carved screen surface.\n\noverall_soundscape:\nA low, continuous indoor room tone sits beneath the faint sputter of the oil lamp wick. The constable's black boots scuff on the stone floor as he steps forward, followed by the creak of the wooden stool and the soft rustle of his dark robe settling as he sits. A faint silk rustle comes from behind the screens.\n\nnon_diegetic_music:\nA sustained low erhu drone at a slow tempo, with a single sparse pipa pluck that enters as the constable sits, gradually increasing in volume before fading without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787708686397,
-    "promptEn": "constable accepts the invitation, moves to sit on a stool closer to the screen",
-    "lastPlayedAt": 1788329345438,
-    "playCount": 21
-  },
-  {
-    "id": "ad611bcbd06b",
-    "source": "h3console",
-    "prompt": "捕快看到女子的手势，松开刀柄走近屏风应了一声",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, with long black hair, a delicate but unnaturally pale face, and a flowing white traditional Chinese robe.\n<Subject 2> is the young male constable in <Picture 3>, with a resolute angular face, a dark long robe, a dao sword at his waist, a brass waist tag reading \"捕\", black boots, and a black headscarf.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens standing in the center, a single oil lamp on a round stone-topped table, a wooden couch bed along the right side, and hanging scrolls on the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> responding to a silent gesture from <Subject 1>. After a beat of hesitation, he releases his grip on the dao hilt and steps toward the carved screens, speaking a line of assent. <Subject 1> watches him from behind the screen gaps as he moves past.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved screens, oil lamp, couch bed, and hanging scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, pale face, and still posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark robe, dao sword, waist tag, boots, headscarf, and resolute face are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber lamplight against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens in the midground, a single oil lamp casting flickering amber light across the round stone-topped table, and deep shadow pools along the walls. <Subject 2>, the young constable in a dark long robe with a dao sword at his waist and a brass waist tag reading \"捕\", stands in the foreground facing the screens, his weight evenly planted on the stone floor. Through the narrow gaps between the screen panels, <Subject 1>, the pale white-robed woman with long black hair, is partially visible; one of her hands slowly extends through a gap, her index finger pointing toward the far corner of the room. The camera pushes in with small amplitude at slow speed. <Subject 2> (S1), the constable, tracks her pointing hand with his eyes while his right hand rests on the dao hilt. He holds his breath for a beat, then releases the hilt, his fingers uncurling slowly, and says in a low, steady voice, <d>[Chinese] 好，我听你的。</d> He steps forward past the oil lamp, his dark robe brushing the air as the small flame sways, sending long shifting shadows across the carved screens.\n[Shot 2] At 00:04.000, the camera cuts to an extreme close-up of <Subject 1>'s eyes seen through a narrow gap between two screen panels. The carved wood frames her gaze on both sides. The amber lamp light catches the surface of her dark iris as she watches <Subject 2> move past, out of frame. Her expression remains perfectly neutral, but a faint, barely perceptible smile crosses her pale lips. The camera holds a static shot. The oil lamp light pulses once, briefly deepening the shadow in her eye socket before returning.\n\noverall_soundscape: The oil lamp wick emits a faint continuous sputter beneath the quiet of the old room. Slow, deliberate boot steps echo off the stone floor as the constable moves past the screens, accompanied by the soft rustle of his dark robe. From behind the screens, a barely audible shift of silk.\n\nnon_diegetic_music: A sustained low erhu drone at a very slow tempo, joined by a single sparse pipa pluck that fades without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787708684658,
-    "promptEn": "constable sees her silent gesture, lets go of his sword hilt, steps closer and agrees",
-    "lastPlayedAt": 1788331070867,
-    "playCount": 22
-  },
-  {
-    "id": "abdbe25f3121",
-    "source": "h3console",
-    "prompt": "捕快识破画皮，狠狠撕开女子的脸，底下是可怕的鬼脸",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the woman in <Picture 2> whose painted face is a beautiful disguise, wearing a flowing white traditional Chinese robe, with pale porcelain skin, long black hair, and delicate features held in an unnaturally still expression.\n<Subject 2> is the young constable in <Picture 3>, with a resolute face, a dark indigo robe, a waist tablet engraved with \"捕\", a dao sword at his belt, black boots, and a black head wrap.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens, a round stone table holding a single lit candle, an arhat bed on the right side, and a hanging scroll on the wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> seeing through <Subject 1>'s painted disguise inside the dim room established by <Picture 1>, and in a burst of rapid motion tearing the painted face apart. He seizes the edge of the painted skin at her cheek and rips it downward in one violent pull as the porcelain features shred like paper, exposing the grotesque demonic visage hidden beneath.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the dim room layout, five carved screens, the candle on the round stone table, the arhat bed, and the hanging scroll are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): partially_preserved - the white robe, long black hair, and pale features are retained until the painted face is torn, after which the painted visage is destroyed to reveal the demon beneath.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the resolute face, dark indigo robe, \"捕\" waist tablet, dao sword, black boots, and black head wrap are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style, with a single warm candle flame against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. Five carved wooden screens stand in the midground beneath deep shadow, a round stone table holds a single lit candle whose warm flame is the only light, an arhat bed occupies the right side, and a hanging scroll hangs on the wall. <Subject 2>, the young constable with a resolute face, dark indigo robe, black head wrap, and \"捕\" waist tablet, a dao sword at his belt, stands in the foreground facing the screens. Through the narrow gaps between the screen panels, <Subject 1>, the woman in the flowing white robe with pale porcelain skin and long black hair, is partially visible, her delicate features held in an unnaturally still expression. The camera holds a static shot with small amplitude shake. <Subject 2>'s eyes narrow, studying her face with growing suspicion. In a single decisive motion he lunges forward, his dark robe sweeping with the speed of his movement, and reaches through the gap to seize the edge of the painted skin at her cheek. The camera pushes in with large amplitude at fast speed. With one sharp, violent pull, he tears the beautiful face downward — the porcelain features split from brow to chin in jagged strips, shredding like dry paper while the white robe and long black hair remain intact. Beneath the shreds, a grotesque, shadowed demonic visage appears, raw and inhuman, its dark hollow eyes fixed on him. The candle flame flings long, swaying shadows across the carved screens, and the torn painted fragments flutter and settle on the stone floor. The shot ends with the exposed demon face staring back through the ruined screen gap as <Subject 2>'s jaw tightens and his grip holds the shredded skin.\n\noverall_soundscape: A sharp, dry fabric-tearing sound rips through the near-silence as the painted skin shreds, followed by the soft patter of torn fragments landing on the stone floor. The candle wick sputters once, and the constable's breathing grows heavier as a faint wind presses against the exterior wall.\n\nnon_diegetic_music: A single low drum strike and a sustained dissonant string swell that cut off abruptly at the instant the painted face tears.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787708674508,
-    "promptEn": "constable sees through her painted disguise and rips her face off, revealing a demonic visage underneath",
-    "lastPlayedAt": 1788327988866,
-    "playCount": 21
-  },
-  {
-    "id": "6d433dcc104c",
-    "source": "h3console",
-    "prompt": "捕快喘着气道谢，靠着榻歇下来，女子在屏风后看着",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, a pale-faced figure in a flowing white traditional Chinese robe with long black hair, delicate features, and an unnaturally still posture, visible through the gaps of the carved screens.\n<Subject 2> is the young constable in <Picture 3>, a man in a dark indigo robe with a dao sword at his waist, a black headwrap, and a brass waist badge reading \"捕\".\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens, a round stone table bearing a single oil lamp, a luohan couch along the right wall, a hanging scroll, and deep shadows on the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> catching his breath inside the dim room established by <Picture 1>. He thanks <Subject 1>, the motionless figure behind the carved screens, then settles onto the couch to rest while the woman watches silently through the screen gaps.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, oil lamp, round stone table, luohan couch, and ambient lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, pale complexion, and still posture are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark indigo robe, dao sword, black headwrap, and brass \"捕\" waist badge are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room: five carved wooden screens stand across the center, a round stone table bearing a single oil lamp sits in the foreground, a luohan couch lines the right wall, a hanging scroll hangs on the far wall, and deep shadow pools across the cracked plaster. <Subject 2>, the young constable in a dark indigo robe with a dao sword at his waist, a black headwrap, and a brass waist badge reading \"捕\", stands just inside the frame, hunched forward with both hands braced on his knees, chest heaving as he catches his breath. The camera holds a static shot while his breathing gradually slows and the sweat on his brow catches the amber light. He lifts his head, straightens his back, and turns slightly toward the carved screens. Through the narrow gaps between the panels, <Subject 1>, the woman in a flowing white robe with long black hair and an unnaturally pale face, stands motionless, her eyes half-shadowed as the amber lamp light catches her cheek. <Subject 2> (S1) says in a low, breathless voice that steadies as he speaks, <d>[Chinese] 多谢你，我先缓一缓。</d> He exhales slowly and lets his shoulders drop. [Shot 2] At 00:04.000, the camera cuts to a medium shot of <Subject 2> settling onto the edge of the luohan couch, his hands resting on his knees and his gaze still fixed on the screens. The camera pushes in with small amplitude at slow speed as the oil lamp flame bends in a faint draft. Behind the screens, <Subject 1>'s white sleeve shifts a fraction and then stills. The constable lowers his eyes with a long, slow exhale, and the amber light steadies across the quiet room.\n\noverall_soundscape:\nRagged breathing gradually steadies into slow exhales while boot soles scrape softly on the stone floor. The oil lamp wick sputters faintly, and a barely audible rustle of silk comes from behind the carved screens before the room settles into quiet.\n\nnon_diegetic_music:\nA low bowed-string drone at a slow tempo that gradually decreases in volume as the constable's breathing steadies, with a single sparse guzheng note plucked near the end that hangs and fades without resolving.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787708663418,
-    "promptEn": "constable catches his breath, thanks her, and rests on the couch while she watches",
-    "lastPlayedAt": 1788326355517,
-    "playCount": 21
-  },
-  {
-    "id": "cd22c2b491fc",
-    "source": "h3console",
-    "prompt": "捕快盯着女子，低声说你继续说，然后不出声等着",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious white-robed woman in <Picture 2>, with a flowing white traditional Chinese robe, pale skin, delicate but unnaturally pale features, and long black hair.\n<Subject 2> is the young male constable in <Picture 3>, with a dark indigo robe, a dao sword at his waist, a brass waist plate engraved with \"捕\", black boots, and a black headband.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens, an oil lamp on a low round stone table, an arhat bed along the right wall, and a hanging scroll on the far wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing still in the dim room established by <Picture 1>, holding the gaze of <Subject 1> through the carved wooden screens. He says in a low voice for her to continue, then falls silent and waits without moving while the oil lamp flame sways.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, oil lamp, round stone table, arhat bed, and hanging scroll are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the white robe, pale skin, delicate features, and long black hair are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark indigo robe, dao sword, waist plate, black boots, and black headband are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp light against deep surrounding shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. Five carved wooden screens stand at the center, their dark frames casting lattice shadows across the stone floor. On the low round stone table before them, a single oil lamp throws flickering amber light that catches the carved patterns; an arhat bed runs along the right wall and a hanging scroll hangs on the far wall, both receding into shadow. <Subject 2>, the young constable in a dark indigo robe with a dao sword at his waist, a brass waist plate engraved with \"捕\", black boots, and a black headband, stands in the foreground, his weight settled evenly and his eyes fixed on the screens. His right hand rests near the dao hilt without drawing; his chest rises and falls with slow, controlled breathing. The camera pushes in with small amplitude at slow speed. Through the narrow gaps between screen panels, <Subject 1>, the white-robed woman with pale skin, delicate features, and long black hair, is partially visible, her face half-lit by the amber glow and her figure unnaturally still. <Subject 2> (S1) says in a low, measured voice, <d>[Chinese] 继续说下去。</d> He closes his lips and stands motionless, waiting. The oil lamp flame bends slightly in the still air, sending a slow wave of shadow across the carved screen surface, while the constable holds the silence without shifting his gaze.\n\noverall_soundscape: The oil lamp wick emits a faint continuous sputter. A subtle rustle of silk comes from behind the screens as the woman shifts slightly. The wooden floorboards creak once under the constable's weight, then near-silence holds with a low draft pressing against the exterior wall.\n\nnon_diegetic_music: A single sustained guzheng harmonic at a very slow tempo with a low bowed-string drone underneath, gradually increasing in volume and holding unresolved.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787708653181,
-    "promptEn": "constable holds her gaze, quietly says \"go on\", then waits in silence",
-    "lastPlayedAt": 1788330366294,
-    "playCount": 22
-  },
-  {
-    "id": "55f98ceec9e4",
-    "source": "h3console",
-    "prompt": "女子从屏风缝里伸手招他，捕快后退一步拒绝",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the woman in <Picture 2>, the pale mysterious woman behind the carved screens, with a delicate face, unnaturally pale white skin, long black hair, and a flowing white traditional Chinese robe.\n<Subject 2> is the constable in <Picture 3>, a resolute young man in a dark long robe with a black headscarf, a dao sword at his waist, and leather boots.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens in the center, a single oil lamp on a round stone table, an arhat bed on the right, and a hanging scroll on the wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2>, the constable, standing in the dim room established by <Picture 1> and facing the carved screens. <Subject 1>, the pale woman behind the screens, extends a hand through a gap as a silent beckoning. <Subject 2> steps back, hand on his dao hilt, and refuses firmly, ending the silent request.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the dim room, five carved screens, oil lamp on the round stone table, arhat bed, and hanging scroll are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the delicate pale face, long black hair, and flowing white robe are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark long robe, black headscarf, dao sword, and resolute bearing are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp key light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. Five carved wooden screens stand across the center, their dark panels picked out only by the single oil lamp on the round stone table, which throws flickering amber light and long shadows across the stone floor. On the right, a low arhat bed and a hanging scroll are barely visible in the gloom. <Subject 2>, the constable in a dark long robe with a black headscarf, a dao sword at his waist, and leather boots, stands in the foreground with his back at a three-quarter angle to the camera, facing the screens. Through the narrow gaps between the screen panels, the pale figure of <Subject 1>, the woman in a flowing white robe with long black hair, is faintly visible. The camera pushes in with small amplitude at slow speed as <Subject 1>'s slender pale hand slowly extends through a gap between two panels, palm turned upward, reaching toward him. <Subject 2>'s jaw tightens; he steps back half a pace, his right hand rising to rest on the dao hilt, his eyes fixed on the reaching hand.\n[Shot 2] At 00:03.000, the camera cuts to a close-up of <Subject 2>'s face from a slightly low angle, the oil lamp casting warm amber light upward across his jaw and cheekbone while the carved screens and the woman's pale silhouette remain softly blurred behind him. The camera holds a static shot. <Subject 2> (S1) says in a firm, low voice: <d>[Chinese] 不行。</d> He closes his mouth into a hard line, his gaze steady and cold. At the bottom edge of frame, his right hand tightens around the dao hilt, knuckles whitening against the dark wrapping.\n\noverall_soundscape: A single slow boot step shifts on the stone floor with a faint leather creak from the sword belt. The oil lamp wick sputters softly. A barely audible rustle of silk comes from behind the screens as the woman's hand withdraws.\n\nnon_diegetic_music: A sustained low dizi drone at a slow tempo with a sparse pipa pluck that increases slightly in volume, then cuts off sharply as the constable refuses.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787708648409,
-    "promptEn": "she reaches a hand through the screen gap beckoning him, he steps back and firmly refuses",
-    "lastPlayedAt": 1788331329079,
-    "playCount": 22
-  },
-  {
-    "id": "3f0c28f5373a",
-    "source": "h3console",
-    "prompt": "捕快隔着屏风打量女子，起了疑心，手悄悄摸向刀柄",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, with long black hair, a pale delicate face, and a flowing white traditional Chinese robe, holding an unnaturally still posture.\n<Subject 2> is the young male constable in <Picture 3>, with a resolute face, a dark long robe, a dao sword at his waist, and a black headscarf.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens, a single oil lamp on a low wooden table, and deep shadows across the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing in the room established by <Picture 1>, studying <Subject 1> through the carved screens as suspicion grows. He notices her unnatural stillness, his hand drifting to the dao hilt while the oil-lamp flame wavers. The two hold a tense, silent standoff through the screen gaps.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved screens, oil lamp position, and deep amber lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the long black hair, pale delicate face, and flowing white robe are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the resolute face, dark long robe, dao sword, and black headscarf are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens standing in the midground, a single oil lamp casting flickering amber light on the low wooden table, and deep shadow pools along the walls. <Subject 2>, the constable in a dark long robe with a dao sword at his waist, stands in the foreground facing the carved screens, his weight still and his jaw set. Through the narrow vertical gaps between the screen panels, <Subject 1>, the woman in a flowing white robe with long black hair, is partially visible, a pale shape lit from one side and standing perfectly motionless. The camera pushes in with small amplitude at slow speed. <Subject 2>'s eyes narrow as he studies her, his gaze traveling from her face to her hands to the hem of her robe. The oil-lamp flame wavers once, throwing long shifting shadows across the carved wood, yet her figure and the folds of her robe do not stir with the air. His right hand drifts to the dao hilt at his waist and closes around the grip without drawing. [Shot 2] At 00:03.500, the shot cuts to an extreme close-up of <Subject 1>'s face seen through a narrow gap between two screen panels. The carved wood frames her gaze on both sides. Her skin is unnaturally pale in the amber light, her eyes half-lidded and still, watching him with a flat, knowing stillness. The camera holds a static shot. <Subject 2>'s shallow breathing is faintly audible, and at the bottom edge of frame his hand remains closed around the sword handle, knuckles tight, unbroken through the final frame.\n\noverall_soundscape: Near-silence. The oil-lamp wick emits a faint continuous sputter. A single quiet creak comes from an old wooden screen frame. Shallow controlled breathing and the low scrape of a hand settling on a sword hilt are faintly audible.\n\nnon_diegetic_music: N/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787708646888,
-    "promptEn": "constable studies her through the screen growing suspicious, hand drifting to his sword",
-    "lastPlayedAt": 1788328925189,
-    "playCount": 21
-  },
-  {
-    "id": "1c0c203cfb3b",
-    "source": "h3console",
-    "prompt": "捕快走到桌边坐下，盯着屏风后女子的表情打量",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens in the midground, a single oil lamp on a round stone table, an arhat bed against the right wall, a hanging scroll on the far wall, and deep shadows along the walls.\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with pale, delicate features, long black hair, and an unnaturally still posture.\n<Subject 2> is the young constable in <Picture 3>, wearing a dark floor-length robe, a sheathed dao sword at his waist, a small waist plate bearing the character \"捕\", a black cloth headwrap, and black boots.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> approaching the lamp-lit table in the room established by <Picture 1>, sitting down on the low stool, and keeping his gaze fixed on <Subject 1> behind the carved screens, carefully studying her expression while the lamp flame sways.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, oil lamp, round stone table, and ambient lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the flowing white robe, long black hair, pale complexion, and still posture are retained throughout.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark robe, sheathed dao, \"捕\" waist plate, black headwrap, and boots are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber lamp light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens in the midground, a round stone table holding a single oil lamp, an arhat bed against the right wall, and a hanging scroll on the far wall. <Subject 2>, the young constable in the dark robe with the sheathed dao at his waist and the \"捕\" waist plate, steps into the frame from the right and walks toward the lamp-lit table. He draws the low wooden stool back with his right hand, lowers himself onto it, settles his weight evenly, and rests his left hand on his knee. His eyes never leave the narrow gaps between the screen panels, where <Subject 1>, the pale woman in the flowing white robe, is partially visible. The camera pushes in with small amplitude at slow speed toward his profile. He studies her expression through the carved gaps: her face is half-lit by the amber lamp glow, beautiful yet unnaturally still, wearing a faint, fixed smile that does not change as the lamp flame sways. His jaw is set, his shoulders stay square, and his breathing remains slow and controlled. The oil lamp flame flickers once, and a long shadow sweeps across the screen panels as the room holds its silence.\n\noverall_soundscape: The oil lamp wick sputters faintly in the close, still air. A wooden stool scrapes against the stone floor, dark fabric rustles as the constable settles, and slow, shallow breathing continues beneath the silence. Distant wind presses faintly against the exterior wall.\n\nnon_diegetic_music: A sustained low bowed-string drone at a slow tempo, with a single sparse pipa pluck landing as the lamp flame flickers, then fading without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787708641782,
-    "promptEn": "constable sits down at the table, studying her expression through the screen",
-    "lastPlayedAt": 1788327636566,
-    "playCount": 21
-  },
-  {
-    "id": "89975be9def4",
-    "source": "h3console",
-    "prompt": "捕快先喝了口水道谢，一边喝一边警惕地看着屏风",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with delicate features, unnaturally pale skin, long black hair, and an unnaturally still posture.\n<Subject 2> is the young male constable in <Picture 3>, wearing a dark robe with a dao sword at his waist, a waist plaque reading \"捕\", black headcloth, black boots, and a resolute, chiseled face.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens standing in the center, a round stone table holding a single oil lamp, a couch bed against the right wall, and hanging scrolls on the cracked plaster wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing in the room established by <Picture 1>, calmly agreeing to drink some water first and thanking <Subject 1>, then stepping to the round stone table, lifting a ceramic cup of water, and drinking while keeping a wary gaze toward the carved screens where <Subject 1> remains motionless.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved screens, round stone table, oil lamp, couch bed, and hanging scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the white robe, long black hair, pale complexion, and still posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark robe, dao sword, \"捕\" name plaque, black headcloth, and boots are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil lamp light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens standing in the midground, a round stone table holding a single oil lamp that casts flickering amber light, a couch bed against the right wall, and hanging scrolls on the cracked plaster wall. <Subject 2>, the young constable in a dark robe with a dao sword at his waist, a waist plaque reading \"捕\", black headcloth, and black boots, stands in the foreground facing the carved screens. The on-screen young man with a resolute face and a calm, low voice (S1) says, <d>[Chinese] 好，我先喝口水，多谢你。</d> He closes his lips, gives a slight nod, then turns and steps toward the round stone table. The camera pushes in with small amplitude at slow speed and settles into a medium close-up as he lifts a ceramic cup of water from the table, raises it to his lips, and takes a slow drink, his eyes staying alert above the cup rim and glancing toward the screens. Behind the screen gaps, <Subject 1>, the pale woman in a flowing white robe with long black hair, stands perfectly motionless, half-lit by the amber oil lamp glow. He lowers the cup as the lamp flame sways, casting long shifting shadows across the carved wooden screens.\n\noverall_soundscape:\nA low indoor room tone with the faint continuous sputter of the oil lamp wick. The constable's dark robe rustles softly as he turns, his boots step across the stone floor, and the ceramic cup clinks gently as it is lifted and set back down. A faint rustle of silk comes from behind the carved screens.\n\nnon_diegetic_music:\nA sustained low erhu drone at a slow tempo with sparse pipa plucks, gradually decreasing in volume before ending with the final frame.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787708626228,
-    "promptEn": "constable drinks some water first, thanks her, watching the screen warily as he drinks",
-    "lastPlayedAt": 1788327006069,
-    "playCount": 21
-  },
-  {
-    "id": "e08f7f7e54dd",
-    "source": "h3console",
-    "prompt": "捕快抱拳道谢，屏风后女子微微点头回应",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the woman in <Picture 2>, a pale-skinned beauty in a flowing white traditional Chinese robe, with long straight black hair and delicate yet bloodless features.\n<Subject 2> is the constable in <Picture 3>, a steely-faced young man in a dark robe with a waist badge stamped \"捕\", a dao sword at his hip, and a black cloth headscarf.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old house with five carved wooden screens in the center, an oil lamp on a round stone table, a luohan bed on the right, and hanging wall scrolls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2>, the constable, standing in the dim room established by <Picture 1>, offering a clasped salute and a single line of gratitude to <Subject 1>, the woman motionless behind the carved screens. She answers with a subtle nod as the oil lamp flame wavers, and the scene settles back into stillness.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved screens, oil lamp, round stone table, luohan bed, and hanging wall scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, pale complexion, and motionless bearing behind the screens are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark robe, waist badge, dao sword, black headscarf, and resolute appearance are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style, lit by a single warm amber oil lamp against deep surrounding shadow, with a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old house: five carved wooden screens stand across the midground, an oil lamp burns on a round stone table in the foreground, a luohan bed sits to the right, and wall scrolls hang in the shadows. In the midground, <Subject 2>, the resolute young constable in a dark robe with a waist badge stamped \"捕\", a dao sword at his hip, and a black cloth headscarf, stands facing the carved screens. The camera pushes in with small amplitude at slow speed. He lifts both hands to his chest in a clasped salute, dips his head forward in a shallow bow, and says in a low, steady voice (S1), <d>[Chinese] 感谢姑娘收留。</d> He holds the pose as the oil lamp flame bends slightly, throwing a slow wave of shadow across the screen surface. Through the gaps between the panels, the pale form of <Subject 1>, the white-robed woman with long black hair, stands perfectly still, half-lit by the amber glow.\n[Shot 2] At 00:04.500, the shot cuts to a close view through a narrow gap between two screen panels, framing <Subject 1>'s face. Her pale, bloodless features catch the amber lamplight as she gives one slow, barely perceptible nod. The camera holds a static shot. The lamp wick sputters once, deepening the shadow across her eye socket for a moment, then the scene settles into stillness.\n\noverall_soundscape:\nA faint room tone persists throughout, with the soft sputter of the oil lamp wick, the subtle rustle of the woman's silk robe shifting behind the screens, and the constable's boots adjusting once on the stone floor.\n\nnon_diegetic_music:\nA sparse sustained guzheng drone at a slow tempo, joined by a single distant woodblock strike, gradually decreasing in volume until it fades to silence in the final second.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787708622306,
-    "promptEn": "constable gives a clasped-hand salute and thanks her, she gives a subtle nod",
-    "lastPlayedAt": 1788330106077,
-    "playCount": 22
-  },
-  {
-    "id": "538dc7e5746f",
-    "source": "h3console",
-    "prompt": "捕快抱拳道谢，屏风后女子微微点头回应",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the woman in <Picture 2>, a pale-skinned beauty in a flowing white traditional Chinese robe, with long straight black hair and delicate yet bloodless features.\n<Subject 2> is the constable in <Picture 3>, a steely-faced young man in a dark robe with a waist badge stamped \"捕\", a dao sword at his hip, and a black cloth headscarf.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old house with five carved wooden screens in the center, an oil lamp on a round stone table, a luohan bed on the right, and hanging wall scrolls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2>, the constable, standing in the dim room established by <Picture 1>, offering a clasped salute and a single line of gratitude to <Subject 1>, the woman motionless behind the carved screens. She answers with a subtle nod as the oil lamp flame wavers, and the scene settles back into stillness.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved screens, oil lamp, round stone table, luohan bed, and hanging wall scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, pale complexion, and motionless bearing behind the screens are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark robe, waist badge, dao sword, black headscarf, and resolute appearance are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style, lit by a single warm amber oil lamp against deep surrounding shadow, with a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old house: five carved wooden screens stand across the midground, an oil lamp burns on a round stone table in the foreground, a luohan bed sits to the right, and wall scrolls hang in the shadows. In the midground, <Subject 2>, the resolute young constable in a dark robe with a waist badge stamped \"捕\", a dao sword at his hip, and a black cloth headscarf, stands facing the carved screens. The camera pushes in with small amplitude at slow speed. He lifts both hands to his chest in a clasped salute, dips his head forward in a shallow bow, and says in a low, steady voice (S1), <d>[Chinese] 感谢姑娘收留。</d> He holds the pose as the oil lamp flame bends slightly, throwing a slow wave of shadow across the screen surface. Through the gaps between the panels, the pale form of <Subject 1>, the white-robed woman with long black hair, stands perfectly still, half-lit by the amber glow.\n[Shot 2] At 00:04.500, the shot cuts to a close view through a narrow gap between two screen panels, framing <Subject 1>'s face. Her pale, bloodless features catch the amber lamplight as she gives one slow, barely perceptible nod. The camera holds a static shot. The lamp wick sputters once, deepening the shadow across her eye socket for a moment, then the scene settles into stillness.\n\noverall_soundscape:\nA faint room tone persists throughout, with the soft sputter of the oil lamp wick, the subtle rustle of the woman's silk robe shifting behind the screens, and the constable's boots adjusting once on the stone floor.\n\nnon_diegetic_music:\nA sparse sustained guzheng drone at a slow tempo, joined by a single distant woodblock strike, gradually decreasing in volume until it fades to silence in the final second.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787708622178,
-    "promptEn": "constable gives a clasped-hand salute and thanks her, she gives a subtle nod",
-    "lastPlayedAt": 1788329161304,
-    "playCount": 21
-  },
-  {
-    "id": "1721c75dbc3b",
-    "source": "h3console",
-    "prompt": "女子从屏风缝里伸出手，让捕快帮她暖手，他犹豫着没碰",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with pale bloodless skin, long black hair, and a delicate, unnaturally still beauty.\n<Subject 2> is the young male constable in <Picture 3>, wearing a dark indigo robe with a dao sword at his waist, a black headscarf, and composed, resolute features.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens at the center, a single oil lamp on a round stone table, an arhat bed along the right wall, and hanging scrolls on the shadowed walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 1> sliding a pale hand through a gap between the carved screens and asking <Subject 2> to warm her hands, while <Subject 2> hesitates with his hand hovering above hers as the oil lamp light flickers.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the dim old room, five carved screens, oil lamp, round stone table, arhat bed, and hanging scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the flowing white robe, long black hair, pale bloodless skin, and unnaturally still beauty are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark indigo robe, dao sword, black headscarf, and resolute bearing are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp light against deep surrounding shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens standing at the center, a single oil lamp on the round stone table casting flickering amber light, an arhat bed along the right wall, and hanging scrolls dimly visible on the cracked plaster walls. <Subject 2>, the constable in a dark indigo robe with a dao sword at his waist, stands in the foreground facing the screens, feet planted and right hand resting near the sword hilt. The camera pushes in with small amplitude at slow speed. Through the narrow vertical gap between two screen panels, <Subject 1>, the woman in a flowing white robe, slides a pale, slender hand into the lamplight, palm upturned and fingers slightly curled, the hand unnaturally still. She speaks in a soft, slow, breathy voice (S1): <d>[Chinese] 那你帮我暖暖手吧。</d> A faint draft bends the oil lamp flame, throwing slow shadows across the carved screens, while her extended hand remains motionless in the amber light.\n[Shot 2] At 00:03.500, the camera cuts to a close-up of the pale hand extended through the screen gap, carved wood framing it on both sides and oil-lamp glow catching the smooth bloodless skin. <Subject 2>'s dark-robed arm enters the bottom edge of frame, his hand hovering an inch above hers, hesitating. The camera holds a static shot as his fingertips tremble slightly and stop, her hand waiting without moving, and the lamp light pulses once across the screen surface.\n\noverall_soundscape: A faint continuous sputter from the oil lamp wick beneath quiet indoor room tone. The soft rustle of white silk as her sleeve slides through the screen gap, followed by a slow creak from the old wooden screen frame, and the subtle shift of the constable's robe as he hesitates.\n\nnon_diegetic_music: A sustained low erhu drone at a slow tempo, joined by a single sparse pipa pluck that lingers without resolving.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787708615415,
-    "promptEn": "she slides her hand through the screen gap asking him to warm it, he hesitates without touching",
-    "lastPlayedAt": 1788328420779,
-    "playCount": 21
-  },
-  {
-    "id": "c2b145b111e2",
-    "source": "h3console",
-    "prompt": "捕快走进屋，注意到屏风后的女子，打了招呼，她只是静静看着",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe with wide sleeves, with pale, unnaturally smooth skin, long black hair falling past her waist, and an expressionless, perfectly still posture.\n<Subject 2> is the young male constable in <Picture 3>, with a stern, angular face, short black hair bound under a black headscarf, a dark indigo robe, a leather waist tag engraved with the character \"捕\", and a dao sword hanging at his waist.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens standing in the center, a single oil lamp on a round stone table, a wooden daybed against the right wall, and a hanging scroll on the wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> stepping into the room established by <Picture 1>, pausing as he notices <Subject 1> standing motionless behind the carved screens, and greeting her. <Subject 1> remains behind the screens and responds only with her silent, watching gaze as the oil lamp flame flickers in the draft.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved wooden screens, oil lamp, round stone table, daybed, and hanging scroll are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, pale complexion, and motionless posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark indigo robe, black headscarf, waist tag, and dao sword are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp light against deep surrounding shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. <Subject 2>, the constable with a stern, angular face, black headscarf, and dark indigo robe, steps through the wooden doorway from the right edge of frame. His dark boots land quietly on the stone floor, his right hand resting on the hilt of the dao sword at his waist, his leather waist tag bearing the character \"捕\" visible in the lamplight. The single oil lamp on the round stone table flickers from the draft, sending long shadows swaying across the five carved wooden screens. He pauses mid-step as his gaze locks onto the pale silhouette of <Subject 1> visible through the gaps between the screen panels. The camera pushes in with small amplitude at slow speed while <Subject 2> straightens his posture, inclines his head slightly in greeting, and speaks in a calm, measured voice (S1), <d>[Chinese] 你好。</d> He closes his lips after the greeting, his eyes remaining fixed on the screen. Behind the carved wood, <Subject 1> stays perfectly still, her white robe barely stirring. [Shot 2] At 00:04.500, the camera cuts to an extreme close-up of <Subject 1>'s eyes seen through a narrow gap between two screen panels, the carved wood framing her face on both sides. Her dark eyes are calm and unnaturally still, watching <Subject 2> from the shadow. The amber lamp light catches the surface of her dark irises, and her lips remain completely closed. A single strand of long black hair shifts across her cheekbone from an imperceptible air current. The camera holds a static shot with a barely perceptible shake. The oil lamp light pulses once, briefly deepening the shadows in her eye sockets before settling back into a gentle sway.\n\noverall_soundscape: A heavy wooden door creaks as it is pushed open, followed by quiet boot steps on stone flooring. The oil lamp wick sputters softly and the flame hisses with each draft. A faint rustle of silk comes from behind the carved screens, then the room settles into near-silence, leaving only the low creak of old wood and the soft sputter of the lamp wick.\n\nnon_diegetic_music: A sustained low guzheng drone at a slow tempo, joined by a single sparse pipa pluck that fades without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787708613364,
-    "promptEn": "constable steps in, notices her behind the screen, greets her, she just watches silently",
-    "lastPlayedAt": 1788331487211,
-    "playCount": 22
-  },
-  {
-    "id": "a5d5d353bbf8",
-    "source": "h3console",
-    "prompt": "捕快举起铜镜，把烛光反射到墙上的美人画像上",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with pale skin, long black hair, and delicate facial features that carry an unnaturally pale, bloodless cast.\n<Subject 2> is the young constable in <Picture 3>, with a resolute, rugged face, wearing a dark robe, a round badge reading \"捕\" at his waist, a dao sword at his belt, black boots, and a black head cloth.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens, a round stone table holding a burning oil lamp and a short candle, a monk's bed on the right, and hanging scrolls along the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> lifting a bronze mirror from the round stone table, angling it to catch the candlelight, and directing the reflected beam toward a beauty portrait hanging on the far wall. The warm light glides across the room and settles on the painted face, illuminating it in sharp detail against the surrounding darkness.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved wooden screens, round stone table, burning lamp and candle, monk's bed, and hanging scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the white robe, long black hair, pale complexion, and silent stillness behind the screens are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark robe, \"捕\" badge, dao sword, and constable appearance are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with a single warm amber light source against deep surrounding darkness and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. Five carved wooden screens stand in the center of the frame; a round stone table holding a burning oil lamp and a short candle sits at their base; a monk's bed occupies the right side; hanging scrolls line the shadowed walls. <Subject 2>, the young constable in a dark robe with a round badge reading \"捕\" at his waist, a dao sword at his belt, black boots, and a black head cloth, stands beside the round stone table, his face set and resolute. Behind the carved screens, the faint pale silhouette of <Subject 1>, the woman in a white robe, is partially visible in the darkness. The camera holds a static shot with small amplitude. The constable leans toward the table, lifts a bronze mirror from the dark surface beside the lamp, and turns it in his hand. He tilts the polished bronze toward the candle flame, and the warm amber light catches the metal, throwing a soft oval of light across the carved screen panels. He rotates his wrist, and the reflected beam glides across the room toward the far wall.\n[Shot 2] At 00:03.000, the camera cuts to a medium close-up of a hanging scroll painting on the shadowed wall — a beauty portrait of a woman in traditional robes, her face rendered in fine, precise brushwork. The reflected amber beam from the mirror, held off-screen, sweeps across the scroll and settles on the painted face. The light holds steady, revealing the painted features in sharp, unnatural clarity against the darkness. The camera pushes in with small amplitude at slow speed. A faint glint of reflected light catches the painted eyes, and the surrounding room remains in deep shadow.\n\noverall_soundscape:\nThe oil lamp flame sputters softly while the candle hisses. The bronze mirror clinks faintly against the stone table as it is lifted, and the constable's dark robe rustles with his movement. Slow footsteps shift on the stone floor, and a low room tone continues underneath.\n\nnon_diegetic_music:\nA low sustained erhu drone at a slow tempo, joined by sparse guzheng plucks that swell gently as the reflected light reaches the beauty portrait, then fade into silence.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674768144,
-    "promptEn": "constable lifts a bronze mirror, reflecting candlelight onto the portrait on the wall",
-    "lastPlayedAt": 1788331188961,
-    "playCount": 22
-  },
-  {
-    "id": "3be1f275fde2",
-    "source": "h3console",
-    "prompt": "捕快拔刀劈开屏风上的封印符，露出后面的女子",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with pale skin, long black hair, and an unnaturally still posture.\n<Subject 2> is the young male constable in <Picture 3>, with a dark robe, a waist badge reading \"捕\", a dao sword at his waist, and a black headscarf.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens in the center, a single oil lamp on a round stone table, an arhat bed along the right wall, and scrolls hanging on the cracked plaster walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> drawing his dao and breaking the sealed central screen panel, tearing through a faded red talisman seal. The split panel reveals <Subject 1> standing motionless behind the broken screen as the oil lamp flame flares from the displaced air.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, oil lamp, round stone table, arhat bed, and wall scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 2]): fully_preserved - the white robe, long black hair, pale complexion, and still posture are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark robe, waist badge, dao sword, and black headscarf are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber lamp light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens standing in the center, a single oil lamp casting flickering amber light across the round stone table, an arhat bed along the right wall, and scrolls hanging in shadow on the cracked plaster walls. <Subject 2>, the young male constable in a dark robe with a waist badge reading \"捕\", a dao sword at his waist, and a black headscarf, stands in the foreground with his weight low and his right hand closed around the dao handle. The camera pushes in with small amplitude at slow speed as he draws the blade in one sharp motion, the steel scraping free with a long metallic ring. He steps twice toward the central screen panel, raises the dao above his shoulder, and swings it in a hard horizontal arc. <Subject 2> (S1) shouts in a fierce, resonant voice, <d>[Chinese] 破开封印！</d> as the blade tears through the faded red talisman seal on the central panel. The carved wood splits apart with a dry splintering crack, sending splinters and a puff of dust into the amber lamplight. The oil lamp flame stretches sideways and flutters violently as the broken panel pitches forward and falls.\n[Shot 2] At 00:04.500, the camera cuts to a closer shot of the shattered gap in the central screen. Through the jagged splintered edges, <Subject 1>, the woman in a flowing white robe with long black hair, stands revealed in the darkness, her pale face half-lit by the guttering oil lamp, her expression perfectly neutral and still. A thin strand of black hair shifts across her cheekbone. The camera holds a static shot as the oil lamp flame settles back to a gentle sway, and the dark silhouette of <Subject 2> with the lowered dao appears at the edge of frame, breathing hard.\n\noverall_soundscape:\nA long metallic ring as the dao clears its scabbard, followed by two heavy boot steps on the stone floor. The blade tears through the carved screen with a dry splintering crack and the shattered panel thuds onto the floor. The oil lamp flame hisses and flutters, and dust patters softly onto the floorboards. Heavy breathing slowly steadies into silence.\n\nnon_diegetic_music:\nA low drum pulse at a fast tempo with a sustained erhu drone, rising in volume and speed toward the strike, then cutting abruptly to near-silence when the woman is revealed, leaving only a faint high harmonic.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674743188,
-    "promptEn": "constable draws his sword and breaks the sealed talisman on the screen, revealing her behind it",
-    "lastPlayedAt": 1788330452357,
-    "playCount": 22
-  },
-  {
-    "id": "693e0b9ca66a",
-    "source": "h3console",
-    "prompt": "捕快慢慢环顾屋子，发现女子正一动不动地隔着屏风看他",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with pale delicate features, long black hair, and an unnaturally still posture.\n<Subject 2> is the young male constable in <Picture 3>, with a resolute face, wearing a dark indigo robe, a black headband, black boots, and a dao sword at his waist beside a wooden token carved \"捕\".\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens, an oil lamp on a round stone table, a luohan bed against the right wall, and hanging scrolls on the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> slowly surveying the dim room established by <Picture 1>, then discovering <Subject 1> watching him motionlessly through the gaps in the carved screens as the oil lamp light shifts around him.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, oil lamp, round stone table, luohan bed, and hanging scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, pale features, and still posture are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark robe, black headband, dao sword, and resolute face are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens standing in the midground, a single oil lamp casting flickering amber light across the round stone table, a luohan bed against the right wall, and hanging scrolls on the cracked plaster walls. <Subject 2>, the young constable in the dark indigo robe with a black headband and a dao sword at his waist, stands in the foreground with his body turned slightly, slowly rotating to survey the room. The camera pans right with small amplitude at slow speed, following his deliberate gaze across the oil lamp, the round stone table, and the luohan bed. The oil lamp flame bends gently, sending long shadows drifting across the carved screen panels. He pauses mid-turn, his head settling toward the screens. Through the narrow vertical gaps between the screen panels, a faint pale shape stands motionless — <Subject 1>, the woman in the white robe with long black hair, her face half-lit by the amber glow. The constable narrows his eyes, studying the still figure behind the carved wood.\n[Shot 2] At 00:03.500, the camera cuts to an extreme close-up of <Subject 1>'s eyes seen through a narrow gap between two screen panels, the carved wood framing her gaze on both sides. The amber lamp light catches the surface of her dark iris. Her expression is perfectly neutral but alert, watching the constable through the gaps. The camera pushes in with small amplitude at slow speed. A single strand of black hair shifts across her pale cheekbone from an imperceptible air current, and the oil lamp light pulses once, briefly deepening the shadow in her eye socket before returning.\n\noverall_soundscape: Near-silence. The oil lamp wick emits a faint continuous sputter, and the constable's dark robe rustles softly as he turns to survey the room. A single quiet creak comes from the old wooden screen frame, and a subtle rustle of silk is heard from behind the screens. Distant wind presses against the exterior wall.\n\nnon_diegetic_music: A sustained low erhu drone at a slow tempo with a sparse guzheng pluck that fades without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674737743,
-    "promptEn": "constable slowly looks around the room, discovers her watching him motionlessly through the screen",
-    "lastPlayedAt": 1788331118906,
-    "playCount": 22
-  },
-  {
-    "id": "c74118d0413f",
-    "source": "h3console",
-    "prompt": "捕快站在桌边看油灯的火苗晃，屏风后女子的影子一动不动",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with pale skin, long black hair, and an unnaturally still posture.\n<Subject 2> is a young male constable in <Picture 3>, wearing a dark long robe, a waist plate reading \"捕\", a dao sword at his waist, black boots, and a black headband, with resolute facial features.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens, an oil lamp on a round stone table, and deep shadows across the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing beside the round stone table in the dim room established by <Picture 1>, observing the oil lamp flame. The flame flickers in a faint draft and the shifting light makes the shadows of the carved screens sway, while <Subject 1> remains a motionless pale silhouette in the gaps behind them.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved wooden screens, oil lamp on the round stone table, and ambient lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, pale complexion, and still posture are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark robe, \"捕\" waist plate, dao sword, black boots, and black headband are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber key light from a single oil lamp against deep surrounding darkness and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium shot of the dim old room with five carved wooden screens in the midground, an oil lamp burning on the round stone table in the foreground, and deep shadow pools along the cracked plaster walls. <Subject 2>, the young constable in a dark long robe with a waist plate reading \"捕\" and a dao sword at his side, stands at the edge of the table, his gaze fixed on the flame. The camera pushes in with small amplitude at slow speed as he inclines his head slightly; the warm amber light rakes across his resolute face and the rough dark fabric of his robe. He lifts his right hand and holds it a few inches above the lamp's glass chimney, palm open, feeling the faint heat. The flame stands nearly still at first, then bends sideways in a barely perceptible draft, and the long shadows of the screen frames sway slowly across the wall. Through the narrow gaps between the carved panels, the faint silhouette of <Subject 1>, the pale woman in a white robe with long black hair, is motionless in the deeper darkness.\n[Shot 2] At 00:04.000, the camera cuts to a close-up of the oil lamp from a low angle, the flame burning bright amber against the surrounding gloom. Beyond the lamp, <Subject 2>'s face is softly lit from below, his brow furrowed, his eyes locked on the flame. The camera holds a static shot. A faint breath of air, as if from behind the screens, makes the flame stretch and gutter, briefly dimming his eyes to shadow before the light settles again. On the polished dark surface of the stone table, the flame's reflection wavers, and in the last gap of the screens behind it, the pale edge of a white sleeve shifts almost imperceptibly, then holds still.\n\noverall_soundscape:\nThe oil lamp wick emits a faint continuous sputter beneath a low indoor room tone. Quiet breathing is audible from the constable, and a single soft creak comes from the old wooden screen frame. Distant wind presses against the exterior wall, and a faint rustle of silk drifts from behind the screens.\n\nnon_diegetic_music:\nA sustained low dizi drone at a slow tempo, joined by sparse guzheng plucks that grow slightly louder and slow down as the flame gutters, then hold on a single unresolved tone.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674716075,
-    "promptEn": "constable watches the oil lamp flame flicker, her silhouette motionless behind the screen",
-    "lastPlayedAt": 1788331090881,
-    "playCount": 22
-  },
-  {
-    "id": "23d1ae1982e6",
-    "source": "h3console",
-    "prompt": "捕快走进屋警惕地打量四周，目光落在屏风后女子的影子上",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with long black hair, an exquisite but unnaturally pale face, and an eerily still posture.\n<Subject 2> is the young constable in <Picture 3>, with a dark indigo robe, a black headcloth, a dao sword at his waist, a brass waist token bearing the character \"捕\", and firm, weathered features.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens, a round stone table bearing a single candle, an arhat bed on the right, and a hanging wall scroll.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2>, the constable, entering the dim old room established by <Picture 1> and surveying his surroundings with cautious attention. His gaze passes over the round stone table, the burning candle, the arhat bed, and the wall scroll before settling on the carved wooden screens, where a faint silhouette of <Subject 1> stands motionless behind the gaps.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, round stone table, candle, arhat bed, and wall scroll are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 2]): fully_preserved - the flowing white robe, long black hair, pale complexion, and still posture are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark indigo robe, black headcloth, dao sword, and constable appearance are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber candlelight against deep black shadows and a slightly desaturated color palette.\n[Shot 1] A medium-wide shot opens from the composition established by <Picture 1>, showing the dim old room with five carved wooden screens standing in the midground, a round stone table bearing a single burning candle on the left, an arhat bed along the right wall, and a hanging wall scroll at the back. <Subject 2>, the constable with a firm, weathered face, wearing a dark indigo robe with a black headcloth and a dao sword at his waist, steps through the doorway and pauses just inside the threshold. The camera holds a static shot with small amplitude shake. He lifts his head, the warm candle glow catching his features, and slowly turns to survey the room. His gaze travels across the round stone table, the flickering candle, the arhat bed, and the wall scroll. The candle flame bends and stretches as he moves, throwing long, shifting shadows across the carved wooden screens.\n[Shot 2] At 00:03.500, the camera cuts to a reverse shot over the constable's shoulder. He stands in the lower foreground, the dark robe and black headcloth filling the frame's edge, facing the carved wooden screens. Through the narrow gaps between the screen panels, a faint pale silhouette is visible — <Subject 1>, the woman in the flowing white robe with long black hair, standing perfectly still. The camera pushes in with small amplitude at slow speed past his shoulder toward the screens. The candle flame flickers, sending a slow wave of shadow across the carved wood, and <Subject 1> remains motionless, her pale face half-lit by the amber glow.\n\noverall_soundscape: A faint draft presses against the old wooden beams while the candle flame sputters softly. Slow boot steps cross the stone floor, and the dark robe rustles as the constable turns. The wooden screen frame creaks once in the silence.\n\nnon_diegetic_music: A low sustained bowed-string drone at a slow tempo, joined by a sparse plucked-string note that rises slightly in pitch and volume before settling into near-silence.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674713724,
-    "promptEn": "constable enters and cautiously looks around, his gaze settling on her silhouette behind the screen",
-    "lastPlayedAt": 1788329847865,
-    "playCount": 22
-  },
-  {
-    "id": "b69389094416",
-    "source": "h3console",
-    "prompt": "捕快站定面对屏风，低声答应会小声说话",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with pale skin, long black hair, and an unnaturally still posture.\n<Subject 2> is the young male constable in <Picture 3>, wearing a dark indigo robe, a black headscarf, black boots, and a dao sword at his waist, with a resolute, alert bearing.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens, a single oil lamp on a round stone table, an arhat bed on the right, and hanging wall scrolls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> stopping in the room established by <Picture 1>, facing the carved screens where <Subject 1> stands motionless. He answers the charged silence with a hushed promise to speak quietly, while the woman remains a still pale presence in the amber lamplight.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, oil lamp, round stone table, arhat bed, wall scrolls, and ambient lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the flowing white robe, long black hair, pale complexion, and unnaturally still posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark indigo robe, black headscarf, black boots, dao sword, and alert bearing are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp light against deep surrounding darkness and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. The camera holds a static shot with small amplitude shake. <Subject 2>, the young constable in a dark indigo robe with a black headscarf, black boots, and a dao sword at his waist, has just stopped near the round stone table; his body is still, his chest rising and falling in controlled shallow breaths, and his right hand rests near the dao hilt. Through the narrow vertical gaps between the five carved screen panels, <Subject 1>, the woman in a flowing white robe with long black hair and pale skin, is partially visible, standing perfectly motionless in the amber-lit dark, her face half-lit by the lamp glow. The single oil lamp on the round stone table casts flickering light across the carved screen surfaces; the arhat bed sits dimly on the right, and the wall scrolls hang still in shadow. The constable's jaw is set and his eyes stay fixed on the pale figure behind the screens as the silence stretches. [Shot 2] At 00:03.500, the camera cuts to a medium close-up of <Subject 2> and pushes in with small amplitude at slow speed. The carved screens and <Subject 1>'s faint white silhouette remain visible, softly out of focus beyond his shoulder. His brow is tense and his lips part slightly. <Subject 2> (S1) says in a low, cautious voice, almost a whisper: <d>[Chinese] 我明白了，我会小声些。</d> He closes his mouth, draws a slow breath, and his gaze flickers briefly toward the dark doorway behind him before returning to the motionless woman behind the screens. The oil lamp flame sways gently, sending slow-moving shadows across his face.\n\noverall_soundscape: A faint continuous sputter from the oil lamp wick and soft creaks of old wood fill the near-silence. The constable's shallow breathing is audible, followed by the quiet rustle of his robe as he settles. Distant wind presses faintly against the exterior wall.\n\nnon_diegetic_music: A sparse, sustained low drone at a very slow tempo with a single soft pipa pluck, gradually fading to near-silence as he speaks.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674695703,
-    "promptEn": "constable stops, faces the screen, quietly promises to keep his voice down",
-    "lastPlayedAt": 1788326721828,
-    "playCount": 21
-  },
-  {
-    "id": "94dc455e26bf",
-    "source": "h3console",
-    "prompt": "捕快答应留下，放下戒备说了句同意的话，女子微微点头",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, a white-robed figure with long black hair falling past her shoulders, unnaturally pale, delicate features, and a still, graceful bearing.\n<Subject 2> is the young constable in <Picture 3>, with a dark robe, a \"捕\" badge at his belt, a dao sword at his waist, black boots, a black headband, and a stern, resolute face.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens, a single oil lamp on a round stone table, an arhat bed on the right, and hanging scrolls along the wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> agreeing to remain with <Subject 1>. He steps deeper into the room established by <Picture 1>, sets his guard aside, and speaks a quiet line of acceptance, while <Subject 1> watches motionless behind the carved wooden screens. The video ends with <Subject 1> giving a barely perceptible nod through the gaps between the panels.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, oil lamp, round stone table, arhat bed, and hanging scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the woman's white robe, long black hair, pale features, and still bearing are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the constable's dark robe, \"捕\" badge, dao sword, black boots, and black headband are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp light against deep surrounding shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens standing in the midground, a single oil lamp casting flickering amber light across the round stone table, and the arhat bed and hanging scrolls receding into darkness along the right wall. <Subject 2>, the young constable in a dark robe with a \"捕\" badge at his belt and a dao sword at his waist, stands just inside the doorway. His chest still rises and falls with short breaths from his earlier running, but his expression has settled into a guarded calm. He turns his head toward the carved screens, holds a beat, then takes one deliberate step into the room. His right hand, which had been curled around the dao hilt, loosens and falls to his side. The camera pushes in with small amplitude at slow speed as he says in a low, steady voice (S1): <d>[Chinese] 好，我陪你待一会儿。</d> He closes his mouth and stands still, his gaze fixed on the narrow gap between two screen panels where <Subject 1>'s white robe is faintly visible in the amber gloom.\n[Shot 2] At 00:03.500, the camera cuts to a close shot of the carved wooden screens from the constable's line of sight. Through the narrow vertical gaps between the panels, <Subject 1>, the woman in a flowing white robe with long black hair and unnaturally pale, delicate features, stands motionless, her face half-lit by the amber lamp glow. She gives a slow, barely perceptible nod. The candle flame on the stone table remains perfectly still. The camera holds a static shot as her faint silhouette settles into the surrounding darkness.\n\noverall_soundscape:\nA faint oil-lamp wick sputter continues throughout the room. Slow, deliberate boot steps on the stone floor are followed by the soft rustle of a dark robe and a single creak from the leather sword belt. Short breaths settle into a slow, even rhythm, while a distant wind presses faintly against the exterior walls.\n\nnon_diegetic_music:\nA sustained low erhu drone at a slow tempo, joined by a single sparse pipa pluck that hangs without resolution. The drone swells faintly as the constable steps forward, then thins out as the woman nods.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674688717,
-    "promptEn": "constable agrees to stay, lets his guard down, she gives a barely-there nod",
-    "lastPlayedAt": 1788328761053,
-    "playCount": 21
-  },
-  {
-    "id": "4874fe3fdbc9",
-    "source": "h3console",
-    "prompt": "捕快举起铜镜狠狠砸在桌上，镜子碎了一地，女子在屏风后看着",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, a pale-faced beauty in a flowing white traditional Chinese robe, with long black hair, an exquisitely delicate but unnaturally pallid face, and a preternaturally still bearing.\n<Subject 2> is the man in <Picture 3>, a young constable in a dark robe with a dao sword at his waist, a brass waist tag reading \"捕\" at his belt, a black headscarf, and black boots, with a resolute, angular face.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens standing at its center, a round stone table bearing an oil lamp, an arhat bed on the right, and hanging scrolls on the wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> lifting a bronze mirror from the stone table, raising it overhead, and smashing it against the tabletop. The mirror shatters into fragments that scatter across the floor while the oil lamp flame lurches and shadows sway across the carved screens. <Subject 1> watches motionless from behind the screens as the shards settle.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved screens, round stone table, oil lamp, arhat bed, and hanging scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the flowing white robe, long black hair, pale face, and motionless bearing behind the screens are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark robe, dao sword, waist tag, black headscarf, black boots, and resolute face are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style, with a single warm amber oil-lamp light against deep surrounding darkness and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. <Subject 2>, the young constable in a dark robe with a brass waist tag reading \"捕\" at his belt and a dao sword at his hip, stands at the left of the round stone table. His right hand grips a bronze mirror resting on the tabletop beside the oil lamp. The camera pushes in with small amplitude at slow speed as he lifts the mirror off the table and raises it high above his shoulder, pausing for a beat while the dull bronze face catches a streak of amber light. He swings the mirror down with force, smashing it against the stone tabletop. The bronze cracks apart with a sharp metallic report; fragments and shards burst outward and scatter across the table and the dark stone floor. The oil lamp flame stretches sideways and flutters violently from the impact, throwing long swaying shadows across the carved wooden screens. Through the narrow gaps between the screen panels, <Subject 1>, the white-robed woman with the pale face, remains perfectly motionless, her dark eyes fixed on the shattered mirror. The shards skitter to a stop and glint in the settling amber light as <Subject 2> straightens over the wreckage, his jaw tight and his breathing heavy.\n\noverall_soundscape: The oil lamp wick sputters faintly beneath the still room tone. Fabric rustles as the bronze mirror is raised, then a sharp metallic crack erupts as it strikes the stone tabletop and shatters. Shards skitter across the stone floor and gradually fall silent, leaving only the low sputter of the lamp wick and a single heavy exhale.\n\nnon_diegetic_music: A low sustained guzheng drone at a slow tempo, with a single muted percussion strike at the moment the mirror shatters, then a long silence as the drone fades out.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674660698,
-    "promptEn": "constable smashes the bronze mirror on the table, it shatters, she watches from behind the screen",
-    "lastPlayedAt": 1788329613656,
-    "playCount": 21
-  },
-  {
-    "id": "ed77b83851a2",
-    "source": "h3console",
-    "prompt": "捕快结印召唤出一只巨大蛤蟆，烟雾中显出身形",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe with wide sleeves, with pale unnaturally white skin, long black hair falling past her waist, and a refined face with a hollow, unnerving stillness.\n<Subject 2> is the young constable in <Picture 3>, wearing a dark indigo robe, a black headscarf, a dark leather waist badge stamped with the character \"捕\", a sheathed dao sword at his belt, black cloth boots, and a resolute, angular face.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens at the center, a round stone table holding a single burning candle, an arhat bed on the right, and hanging scrolls along the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing in the room established by <Picture 1>, rapidly forming hand seals, and summoning a colossal toad through a burst of smoke, while <Subject 1> remains motionless behind the carved screens. The candlelight flickers as the toad materializes and looms over the room.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, round stone table, candle, arhat bed, hanging scrolls, and ambient lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the white robe, long black hair, pale complexion, and motionless posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark indigo robe, black headscarf, \"捕\" waist badge, dao sword, and resolute face are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm flickering candlelight against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. <Subject 2>, the young constable in the dark indigo robe with the black headscarf, \"捕\" waist badge, and dao at his belt, stands in the midground facing the five carved screens, his weight planted on the stone floor. Behind the narrow gaps between the screen panels, <Subject 1>, the white-robed woman with long black hair and pale skin, is faintly visible, utterly still. The camera pushes in with small amplitude at slow speed as <Subject 2> raises both hands to chest height and begins forming hand seals — his fingers snap through a rapid sequence of interlocked gestures, wrists crossing, fingers weaving, palms rotating, each gesture snapping crisply to the next. The single candle on the stone table throws long wavering shadows across the screens and walls with every movement. He ends the sequence by pressing both palms flat together, fingers interlaced, jaw set.\n[Shot 2] At 00:04.000, the camera cuts to a wider shot of the room. A dense white burst of smoke erupts from the floor in front of <Subject 2>, rolling outward across the stone floor. From within the smoke a colossal toad materializes, rising to fill the frame — mottled dark-green skin, a wide scarred jaw, heavy-lidded eyes, and a long smoking pipe clenched between its teeth, its bulk dwarfing the carved screens and the stone table. The candle flame stretches and gutters as the toad settles, dust drifting down from the ceiling beams, while <Subject 2> stands in the foreground with his palms still pressed together and the smoke curling around his dark robes.\n\noverall_soundscape:\nThe candle wick sputters softly throughout. The constable's robe rustles and his palms snap together with each hand seal. A low subterranean rumble swells as the smoke erupts, followed by a heavy muffled thud as the toad lands, old floorboards creaking and dust sifting down.\n\nnon_diegetic_music:\nA low sustained cello drone at a slow tempo with sparse taiko drum hits accenting each hand-seal gesture, swelling abruptly at the smoke burst and landing on a single deep boom as the toad materializes, then trailing into low rumbling silence.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674642139,
-    "promptEn": "constable forms hand seals and summons a giant toad out of smoke",
-    "lastPlayedAt": 1788331343089,
-    "playCount": 22
-  },
-  {
-    "id": "e9a6320e18a9",
-    "source": "h3console",
-    "prompt": "捕快从旁边进来，坐在油灯边的小凳上歇着",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, with unnaturally pale skin, delicate refined features, long black hair, and a flowing white traditional Chinese robe.\n<Subject 2> is the young male constable in <Picture 3>, with a resolute weathered face, a dark indigo robe, a black headscarf, black boots, a dao sword at his waist, and a small bronze waist tablet engraved with the character \"捕\".\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens across the center, an oil lamp on a low wooden table, a round stone table holding an unlit candle, and an arhat bed against the right wall beneath a hanging scroll.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> entering from the right of the room established by <Picture 1>, lowering himself onto the low stool beside the oil lamp to rest, and settling as the lamp flame wavers. <Subject 1> remains a faint, still presence visible through the gaps between the carved screens.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, oil lamp, round stone table, and arhat bed are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 2]): partially_preserved - the flowing white robe and long black hair remain visible through the screen gaps while her face stays mostly concealed in shadow.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark indigo robe, black headscarf, dao sword, black boots, and waist tablet are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp light against deep surrounding darkness and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. The oil lamp casts flickering amber light across the low wooden table, a round stone table holding an unlit candle stands in the midground, an arhat bed rests against the right wall beneath a hanging scroll, and long shadows pool beneath the five carved wooden screens. The camera holds a static shot with small amplitude shake. From the right edge of frame, <Subject 2>, the young constable in a dark indigo robe with a dao sword at his waist, steps quietly into the room. He crosses to the low stool beside the oil lamp, turns, and lowers himself onto the seat, his dark robe settling around him as he places both hands on his knees. The lamp flame bends slightly from the movement of his robe, then recovers.\n[Shot 2] At 00:03.500, the camera cuts to a close-up of <Subject 2> seated beside the oil lamp. Warm amber light flickers across his jaw and cheekbone while deep shadow fills the rest of the frame. He lets out a slow breath, his shoulders easing. His gaze drifts toward the carved screens; through a narrow gap between panels, the pale white robe of <Subject 1> is faintly visible, perfectly motionless. The camera pushes in with small amplitude at slow speed. The small bronze waist tablet engraved with \"捕\" catches the lamp light at his belt. The lamp flame wavers once, briefly deepening the shadow in his eye sockets, and he holds his position as the room settles into quiet.\n\noverall_soundscape: Slow deliberate footsteps on stone flooring, followed by the soft rustle of a dark robe as he sits. The oil lamp wick sputters faintly, the low wooden stool creaks once, and a slow breath steadies. A faint wind presses against the old walls from outside.\n\nnon_diegetic_music: A sustained low erhu drone at a slow tempo, joined by a single sparse guzheng pluck that fades without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674642026,
-    "promptEn": "constable enters from the side and sits down on a stool by the lamp to rest",
-    "lastPlayedAt": 1788330252201,
-    "playCount": 22
-  },
-  {
-    "id": "fe7ffbe65653",
-    "source": "h3console",
-    "prompt": "捕快举铜镜照自己，镜子里女子出现在肩后，转身却发现屏风空的",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with a beautiful but unnaturally pale face, long black hair, and an unnaturally still posture.\n<Subject 2> is the young constable in <Picture 3>, with a resolute face, wearing a dark robe with a waist plaque reading \"捕\", a dao sword at his waist, black boots, and a black headcloth.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens in the midground, a round stone table holding a candle and a bronze mirror, a couch bed on the right, and hanging wall scrolls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> crossing the dim old room established by <Picture 1>, lifting the bronze mirror from the round stone table, and examining his reflection in the candlelight. In the mirror's dark surface, <Subject 1> appears reflected just behind his shoulder. When he turns around, the room behind him is empty; the carved screens stand still with <Subject 1> motionless behind them.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, round stone table with the candle and bronze mirror, couch bed, and wall scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2], [Shot 3]): fully_preserved - the white robe, long black hair, pale skin, and unnaturally still posture are retained throughout.\n<Subject 2> (appears in [Shot 1], [Shot 2], [Shot 3]): fully_preserved - the dark robe, waist plaque, dao sword, black headcloth, and resolute face are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber candlelight against deep surrounding shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room: five carved wooden screens stand in the midground, a round stone table holds a single candle and a round bronze mirror, a couch bed sits at the right edge, and hanging scrolls line the far wall. <Subject 2>, the constable with a resolute face wearing a dark robe, a waist plaque reading \"捕\", a dao sword at his waist, black boots, and a black headcloth, steps into frame from the left. The camera tracks right with small amplitude at slow speed as he crosses toward the table. Behind the carved screens, <Subject 1>, the woman in a flowing white robe with long black hair, is a faint, motionless silhouette visible through the gaps. <Subject 2> reaches the round stone table, his fingers close around the bronze mirror, and he lifts it from the stone surface, the candle flame glinting across the polished metal.\n[Shot 2] At 00:04.000, the camera cuts to a close-up of the round bronze mirror held before <Subject 2>'s face. The dark polished surface catches the trembling candle flame, and his own reflected face stares back, half-lit by amber light. Yet just behind his reflected shoulder, the pale face of <Subject 1> stands in the bronze — unnaturally still, white robe hanging motionless — while the reflection of the room behind him shows only empty shadow. The camera pushes in with small amplitude at slow speed toward the mirror, and the reflected woman holds perfectly still without moving a single strand of hair.\n[Shot 3] At 00:07.000, the camera cuts to a wide shot from behind <Subject 2> as he snaps his head around, the dao at his waist shifting with the motion. The space behind him is empty: the candle burns steadily, the couch bed stands bare, and no one crosses between the five carved screens. Through the narrow gaps, <Subject 1> remains a faint white silhouette, motionless in the same pose as before. The camera holds a static shot with small amplitude shake. <Subject 2> exhales slowly and lowers the mirror to his side, his jaw set, his eyes still fixed on the screens.\n\noverall_soundscape: The candle flame hisses softly and the old wooden screens emit a faint creak. Boot steps cross the stone floor and stop. The bronze mirror scrapes lightly against the stone tabletop as it is lifted, followed by a long silence. In the final beat, a single sharp intake of breath, then stillness.\n\nnon_diegetic_music: A sustained low erhu drone at a slow tempo, joined by a sparse guzheng pluck that swells as the reflection appears and cuts off abruptly when <Subject 2> turns around.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674639055,
-    "promptEn": "constable checks his reflection in the mirror, sees her behind his shoulder in it, turns around to find the screen empty",
-    "lastPlayedAt": 1788326341499,
-    "playCount": 21
-  },
-  {
-    "id": "ebaf0dde3d9f",
-    "source": "h3console",
-    "prompt": "捕快对着屏风鞠躬，谢谢女子收留他",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with pale skin, long black hair, and an unnaturally still posture.\n<Subject 2> is the young male constable in <Picture 3>, wearing a dark robe with a waist tablet reading \"捕\", a dao sword at his waist, black boots, and a black headcloth.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens in the center, a round stone table with a single candle, a luohan bed against the right wall, and aged hanging scrolls on the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing in the room established by <Picture 1>, offering his gratitude to <Subject 1> for sheltering him. He bows formally and speaks the line while <Subject 1> remains a motionless pale silhouette behind the carved wooden screens, watching him through the gaps.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved wooden screens, round stone table, candle, luohan bed, and hanging wall scrolls are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the flowing white robe, long black hair, pale complexion, and still posture are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark robe, waist tablet reading \"捕\", dao sword, black boots, and black headcloth are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber candlelight against deep surrounding shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium shot of the dim old room. Five carved wooden screens stand in the center, a round stone table with a single flickering candle sits in the lower foreground, a luohan bed is visible against the right wall, and faded hanging scrolls line the walls. <Subject 2>, the young constable in a dark robe with a waist tablet reading \"捕\", a dao sword at his waist, black boots, and a black headcloth, stands in the foreground facing the screens. The camera holds a static shot with small amplitude shake. He draws his hands together in a formal salute and inclines his head. <Subject 2> (S1) says in a measured young male voice with a respectful tone: <d>[Chinese] 我明白，多谢你收留我。</d> As he finishes, the candle flame wavers, sending a long shifting shadow of his figure across the carved screens. Through the gaps between the screen panels, the pale silhouette of <Subject 1>, the woman in a white robe with long black hair, stands motionless, half-lit by the warm amber glow.\n[Shot 2] At 00:03.500, the camera cuts to a close-up of <Subject 2>'s face from a frontal angle. Candlelight traces the line of his jaw and the black headcloth as he slowly lifts his gaze from the flame toward the screens. His brow is set, his expression solemn and trusting. The camera pushes in with small amplitude at slow speed. In the background beyond his shoulder, through the dark gaps between the screen panels, <Subject 1>'s white form shifts almost imperceptibly, her head tilting a few degrees to one side, watching him. The candle flame steadies and the room settles back into stillness.\n\noverall_soundscape: A faint continuous sputter from the candle wick sits beneath low room ambience. The constable's robe rustles softly as he draws his hands together in salute, and a subtle silk shift comes from behind the carved screens.\n\nnon_diegetic_music: A low sustained erhu drone at a slow tempo with sparse pipa plucks, gradually decreasing in volume and fading without resolution.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674623678,
-    "promptEn": "constable bows to the screen, thanking her for sheltering him",
-    "lastPlayedAt": 1788327442438,
-    "playCount": 21
-  },
-  {
-    "id": "d94991a1e21b",
-    "source": "h3console",
-    "prompt": "捕快隔着屏风缝盯着女子，镜头慢慢推近她的脸",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens standing across the center, a round stone table holding a single lit candle, a wooden arhat bed against the right wall, and a hanging scroll on the far wall.\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with long black hair, an unnaturally pale face, and a delicate, slightly hollow expression.\n<Subject 2> is the young male constable in <Picture 3>, in a dark indigo robe with a waist badge reading \"捕\", a dao sword at his hip, a black headscarf, and a resolute, weathered face.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing in the room established by <Picture 1>, silently observing <Subject 1> through the gaps of the carved screens. Her unnaturally pale face and absolute stillness hold his gaze as the candlelight wavers, and the camera closes in until her painted-smooth face fills the frame.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the carved screens, round stone table, candle, arhat bed, and hanging scroll are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, and pale delicate face are retained throughout.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark indigo robe, waist badge, dao sword, black headscarf, and resolute face are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber candlelight against deep surrounding darkness and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. Five carved wooden screens stand across the center, and a single candle on the round stone table casts flickering amber light that barely reaches the walls, leaving deep shadow pools around the arhat bed on the right and the hanging scroll beyond it. <Subject 2>, the young constable in a dark indigo robe with a dao sword at his hip, stands in the foreground facing the screens, his body still and his gaze fixed ahead. The camera pushes in with small amplitude at slow speed, passing his shoulder as he watches. Through the narrow vertical gaps between the carved panels, <Subject 1>, the woman in a flowing white robe with long black hair, is partially visible. She stands perfectly motionless in the dim amber light, her pale face half-lit, her hands folded at her waist, her eyes open and fixed straight ahead. The candle flame bends once, sending a slow wave of shadow across the carved wood. <Subject 2> does not move; his right hand drifts to the hilt of his dao, fingers resting on the wrapping without drawing the blade.\n[Shot 2] At 00:03.500, the camera cuts to an extreme close-up of <Subject 1>'s face seen through a narrow gap between two screen panels, the carved wood framing her on both sides. The amber candlelight catches the surface of her dark iris and the unnaturally smooth, too-perfect texture of her pale skin, which has the flat, finished quality of a painted portrait. Her expression is serene and empty, lips slightly parted but perfectly still. A single strand of black hair shifts across her cheekbone from an imperceptible air current. The camera holds a static shot. In the faint highlight of her dark eye, the candle flame flickers at a distance, and the faint silhouette of <Subject 2> slides across the lower surface of her iris, still watching.\n\noverall_soundscape: The candle wick emits a faint continuous sputter while low room tone holds underneath. A single quiet creak comes from the old wooden screen frame, followed by the soft rustle of silk from behind the screens. Distant wind presses against the exterior wall.\n\nnon_diegetic_music: A sustained low guzheng harmonic at a very slow tempo with a deep bowed-string drone underneath, gradually increasing in volume before cutting off abruptly at the final frame.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674609636,
-    "promptEn": "constable silently watches her through the screen gaps, camera slowly zooming in on her face",
-    "lastPlayedAt": 1788329901901,
-    "playCount": 22
-  },
-  {
-    "id": "5e53f7f5fe9b",
-    "source": "h3console",
-    "prompt": "玩家慌张闯进屋，回头喘着气说外面有东西在追他",
-    "fullPrompt": "玩家慌张地闯入房间，回头望向门外，喘着气说\"对不起，我不是有意闯进来的，外面有东西在追我。\" <Picture 1> 场景为参照，<Picture 2> NPC 表情惊讶回应。",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674446841,
-    "promptEn": "player panics into the room, glances back, out of breath, says something's chasing him outside",
-    "lastPlayedAt": 1788329637674,
-    "playCount": 21
-  },
-  {
-    "id": "c25d53a5e086",
-    "source": "h3console",
-    "prompt": "捕快隔着屏风盯着女子，两人都不说话，气氛很紧张",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with an unnaturally pale face, long black hair falling past her shoulders, and an unnaturally still, expectant posture.\n<Subject 2> is the young male constable in <Picture 3>, with a dark robe, a dao sword at his waist, black boots, a black headscarf, and a resolute, composed expression.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens in the center, a single oil lamp on a low round stone table, an arhat bed on the right, and a hanging scroll on the far wall.\n\nsummary:\n[keyframe completion + reference generation] The target video captures a tense silence as <Subject 2> holds the gaze of <Subject 1> without speaking. The camera slowly pushes in on his face while the oil lamp flame wavers, then cuts to <Subject 1>'s eyes watching him through the gaps in the carved screens. Neither figure moves or speaks.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved wooden screens, oil lamp, round stone table, arhat bed, hanging scroll, and low amber lighting are retained.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, pale complexion, and motionless bearing are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark robe, dao sword, black boots, black headscarf, and composed posture are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with a warm single-source oil lamp against deep surrounding darkness and extensive negative space.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium shot of the dim old room. Five carved wooden screens stand in the center, their lattice panels casting long angular shadows under the low amber light of a single oil lamp on a round stone table; a hanging scroll and a low arhat bed sit against the far wall, half-lost in shadow. <Subject 2>, the constable in a dark robe with a dao sword at his waist, black boots, and a black headscarf, stands in the foreground facing the screens, his weight evenly planted on the stone floor and his arms hanging motionless at his sides. Through the narrow vertical gaps between the screen panels, <Subject 1>, the woman in a flowing white robe with long black hair, is partially visible, her unnaturally pale face half-lit by the amber glow and her dark eyes fixed on him. The camera pushes in with small amplitude at slow speed toward <Subject 2>'s profile. His jaw is set, his eyes steady on the woman, his chest barely rising with controlled breathing. The oil lamp flame bends slightly, sending a slow wave of shadow across the carved screen surface. Neither figure moves. The silence holds. [Shot 2] At 00:03.500, the camera cuts to an extreme close-up of <Subject 1>'s eyes seen through a narrow gap between two screen panels. The carved wood frames her gaze on both sides. The amber lamp light catches the surface of her dark iris, and her expression is perfectly neutral but alert, waiting. A single strand of black hair shifts across her cheekbone from an imperceptible air current. The camera holds a static shot. The oil lamp light pulses once, briefly deepening the shadow in her eye socket before returning.\n\noverall_soundscape: Near-silence. The oil lamp wick emits a faint continuous sputter, and a single quiet creak comes from the old wooden screen frame. Distant wind presses against the exterior wall.\n\nnon_diegetic_music: N/A",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674395006,
-    "promptEn": "constable and the woman hold a tense silent stare through the screen",
-    "lastPlayedAt": 1788329329423,
-    "playCount": 21
-  },
-  {
-    "id": "f264bce22903",
-    "source": "h3console",
-    "prompt": "捕快按着胸口发誓要把心交给女子，她露出意味深长的笑",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with delicate pale features, long black hair, and an unnaturally still bearing.\n<Subject 2> is the young male constable in <Picture 3>, with a dark robe, a black headwrap, a dao sword at his waist, a wooden waist plaque bearing the character \"捕\", and a stern, resolute face.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens across the center, a single oil lamp on the round stone table, a wooden daybed along the right wall, and hanging scrolls on the shadowed plaster walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing before the carved screens, pressing his hand over his heart and vowing in a low solemn voice to give his heart to <Subject 1>, the pale woman behind the screens. The camera then closes on <Subject 1>'s face as her lips curve into a faint knowing smile.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, round stone table, oil lamp, daybed, wall hangings, and amber shadow lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the flowing white robe, long black hair, pale complexion, and still bearing are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark robe, black headwrap, dao sword, waist plaque bearing \"捕\", and stern resolute face are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber lamp light against deep surrounding darkness and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room: five carved wooden screens stand across the center, a single oil lamp burns on the round stone table, a wooden daybed lines the right wall, and hanging scrolls catch faint amber light on the shadowed plaster walls. <Subject 2>, the young constable in a dark robe with a black headwrap, a dao sword at his waist, and a wooden waist plaque bearing the character \"捕\", stands facing the screens with his weight evenly planted on the stone floor. Through the narrow vertical gaps between the screen panels, <Subject 1>, the woman in a flowing white robe with pale skin and long black hair, is partially visible, her face half-lit by the amber glow. The camera pushes in with small amplitude at slow speed toward <Subject 2>. He draws a slow, steady breath, then raises his right hand and presses it flat against his chest over his heart. In a low, solemn voice, he says, <d>[Chinese] 我愿意把心交给你。</d> He holds his hand over his heart, his eyes steady on the woman's silhouette, while the oil lamp flame bends slightly and a slow wave of shadow drifts across the carved screen surface.\n[Shot 2] At 00:04.000, the camera cuts to a close-up of <Subject 1>'s face seen through a narrow gap between two screen panels. The carved wood frames her pale face on both sides, and the amber lamp light catches the surface of her dark eyes. Her lips curl into a faint, almost imperceptible smile while her gaze remains fixed on <Subject 2>. A single strand of black hair shifts across her cheekbone from an imperceptible air current. The oil lamp flame pulses once, briefly deepening the shadows in her eye sockets before settling. The camera holds a static shot.\n\noverall_soundscape:\nThe oil lamp wick sputters faintly and continuously, joined by a single quiet creak from the old wooden screen frame. The constable's dark robe rustles softly as he moves, and his slow, steady breathing is audible. Distant wind presses against the exterior walls.\n\nnon_diegetic_music:\nA low sustained erhu drone at a slow tempo with widely spaced pipa plucks, gradually increasing in volume during the second shot and holding steady through the final frame.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674361187,
-    "promptEn": "constable presses his hand to his chest, vowing his heart to her, she gives a knowing smile",
-    "lastPlayedAt": 1788330396320,
-    "playCount": 22
-  },
-  {
-    "id": "8aba0a6bf05b",
-    "source": "h3console",
-    "prompt": "捕快半夜査古宅，烛光摇曳，女的躲在屏风后",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the pale-skinned woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with long black hair, refined but unnaturally pale features, and a still, waiting posture.\n<Subject 2> is the young constable in <Picture 3>, wearing a dark long robe, a black headscarf, a dao sword at his waist, and dark boots, with a resolute, guarded face.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens in the center, a round stone table holding a single oil lamp, an arhat bed on the right, and a hanging scroll on the wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> following the woman's instruction to lower his voice, then quietly scanning the room: first the closed door and latticed window, then the oil lamp on the round stone table, and finally <Subject 1>'s position behind the carved screens. His guarded observation ends with the lamp flame wavering and <Subject 1>'s pale silhouette holding still in the dark.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, five carved screens, round stone table, oil lamp, arhat bed, hanging scroll, and dim amber lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, pale complexion, and still posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark long robe, black headscarf, dao sword, dark boots, and guarded bearing are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber oil-lamp light against deep shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room: five carved wooden screens stand in the center, the round stone table with a single oil lamp sits at frame left, an arhat bed is visible on the right, and a hanging scroll darkens the far wall. <Subject 2>, the young constable in a dark long robe with a black headscarf and a dao sword at his waist, stands in the foreground, his body angled toward the screens. Heeding the woman's words, he deliberately lowers his voice, his words dropping to a hushed murmur, and his breathing slows. The camera pushes in with small amplitude at slow speed as he begins to observe. First his gaze flicks to the closed door and latticed window at frame left, checking the exit. Then his eyes move to the oil lamp on the round stone table, watching its flame bend slightly as if in an unseen draft, the amber glow flickering across the carved screen surfaces. His right hand drifts to rest near the dao hilt. [Shot 2] At 00:03.500, the camera cuts to a medium close-up over <Subject 2>'s shoulder, his dark robe and black headscarf filling the lower-left of the frame, his head still and his eyes visible in profile. Through the narrow gaps between two screen panels ahead, <Subject 1>, the white-robed woman with long black hair and pale skin, stands motionless in the amber-lit dark. The oil lamp at frame right throws her pale figure into faint relief while the shadow between the panels deepens. The camera holds a static shot as the lamp flame trembles once and her silhouette remains perfectly still.\n\noverall_soundscape: A faint draft rustles the hanging scroll and stirs a low hiss from the oil lamp flame. The constable's dark robe whispers as he moves with quiet, deliberate care; a brief low male murmur fades as he falls silent. His steps on the stone floor are slow and soft, while distant wind presses against the old wood outside.\n\nnon_diegetic_music: A sustained low bowed-string drone at a slow tempo, joined by a sparse, widely spaced plucked note repeating at irregular intervals, gradually increasing in volume as the shot settles on the woman, then cutting off abruptly.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787674351536,
-    "promptEn": "constable searches an old house at night, candle flickering, woman hiding behind the screen",
-    "lastPlayedAt": 1788331281036,
-    "playCount": 22
-  },
-  {
-    "id": "dc13e4f165dc",
-    "source": "h3console",
-    "prompt": "捕快小心跟着女子的手势走，手搭在刀柄上盯着她",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with long black hair, pale skin, a delicate but unnaturally pallid face, and an eerily still posture.\n<Subject 2> is the young constable in <Picture 3>, with a resolute face, wearing a dark robe with a waist tag reading \"捕\", a dao sword at his waist, black boots, and a black headscarf.\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens in the center, a single candle on a round stone table, an arhat bed on the right, and scrolls hanging on the wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> cautiously following a silent gesture from <Subject 1>. He steps forward as directed but keeps his right hand near the dao hilt, eyes tracking the woman's movement behind the screens. The candle flame shifts as he moves through the room.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room geometry, carved screens, candle position, arhat bed, wall scrolls, and lighting conditions are retained.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the white robe, long black hair, pale complexion, and composed demeanor are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the dark robe, waist tag \"捕\", dao sword, black boots, headscarf, and alert bearing are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber candlelight against deep surrounding darkness and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium shot of the dim old room. Five carved wooden screens stand in the center, a single candle flame flickers on the round stone table, an arhat bed sits on the right, and scrolls hang on the wall. <Subject 2>, the constable in a dark robe with a resolute face, stands in the foreground facing the screens. Behind the screen gaps, <Subject 1>, the woman in a flowing white robe with long black hair, raises one pale hand slowly and extends her index finger, pointing toward the far corner of the room. The camera trucks right with small amplitude at slow speed, following <Subject 2> as he takes one deliberate step in the indicated direction. His left foot slides forward across the stone floor while his right hand drifts to the dao hilt at his waist, fingers curling around the grip without drawing. His waist tag \"捕\" catches the candlelight for a moment. His eyes remain locked on the gaps between the screen panels where <Subject 1>'s white sleeve is visible. [Shot 2] At 00:03.000, the camera cuts to a close-up of <Subject 2>'s face from a low angle. The candlelight casts upward amber light across his jaw and cheekbone while his eyes stay fixed on a point beyond frame left where <Subject 1> stands. His brow is tight, lips pressed together. The camera pushes in with small amplitude at slow speed. A faint reflection of <Subject 1>'s white robe slides across the dark surface of his iris. His right hand, visible at the bottom edge of frame, tightens its grip on the dao handle, knuckles whitening against the dark hilt wrapping.\n\noverall_soundscape:\nSlow, measured boot steps on stone with a faint leather creak from the sword belt. The candle wick sputters once. A subtle rustle of silk comes from behind the screens.\n\nnon_diegetic_music:\nA single sustained guzheng harmonic at a very slow tempo, with a low bowed string drone underneath that gradually increases in volume before cutting off abruptly.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787672849330,
-    "promptEn": "constable cautiously follows her silent gesture, hand near his sword, tracking her movement",
-    "lastPlayedAt": 1788326499631,
-    "playCount": 21
-  },
-  {
-    "id": "feb4df69e1a2",
-    "source": "h3console",
-    "prompt": "捕快隔着屏风客气地向女子讨茶喝",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe, with long black hair and an exquisite but unnaturally pale face.\n<Subject 2> is the young male constable in <Picture 3>, wearing a dark long robe, with a black headwrap, black boots, a dao sword at his waist, and a bronze waist badge inscribed with \"捕\".\n<Picture 1> is the first frame of [Shot 1], establishing the dim old room with five carved wooden screens, a round stone table holding a candle, a low daybed on the right, and hanging scrolls on the wall.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> standing in the dim room established by <Picture 1>, politely requesting tea from <Subject 1> behind the carved screens. <Subject 1> remains motionless, then a pale hand extends through the screen gap holding a steaming white cup, which <Subject 2> hesitantly reaches for.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout with five carved screens, round stone table, candle flame, low daybed, and hanging scrolls is retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1], [Shot 2]): fully_preserved - the woman's white robe, long black hair, and pale complexion are retained.\n<Subject 2> (appears in [Shot 1], [Shot 2]): fully_preserved - the constable's dark robe, black headwrap, dao sword, and the \"捕\" waist badge are retained.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with a single warm candle flame cutting through deep shadow and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room. Five carved wooden screens stand in the midground, their dark lacquered panels catching faint amber glints. A round stone table in the foreground holds a single burning candle, the only source of light, casting a small warm pool on the stone surface and throwing long, wavering shadows of the screen frames across the cracked plaster walls. A low daybed sits along the right wall beneath a hanging scroll. <Subject 2>, the young constable from <Picture 3> with a dark long robe, a black headwrap, black boots, a dao sword at his waist, and a bronze waist badge inscribed with \"捕\", stands beside the stone table facing the screens, hands clasped loosely in front of him. The camera holds a static shot with small amplitude shake. He draws a slow breath and, in a calm steady young male voice, <Subject 2> (S1) says: <d>[Chinese] 请给我倒杯茶.</d> Behind the screens, <Subject 1>, the pale woman in a flowing white robe with long black hair, is faintly visible through the gaps between panels, standing perfectly motionless with her face half-lost in shadow. The candle flame bends once, as if stirred by an unseen draft, and the shadows across the screens shift. He waits, his eyes fixed on the screen gaps.\n[Shot 2] At 00:03.500, the camera cuts to a close-up of the narrow gap between two screen panels. From the darkness behind the screens, a slender, unnaturally pale hand belonging to <Subject 1> emerges, holding a white porcelain cup filled with gently steaming tea. The hand extends through the gap with a smooth, unhurried motion and sets the cup down on the edge of the round stone table with a soft click. The candle flame bends sideways as the hand passes above it, then settles back. At the lower edge of frame, <Subject 2>'s hand pauses, hovering above the cup for a beat, then reaches slowly forward until his fingertips brush the warm porcelain. The camera pushes in with small amplitude at slow speed as the pale hand withdraws back into the darkness between the panels, leaving the steaming cup on the table.\n\noverall_soundscape: The candle flame sputters softly. Silk rustles faintly from behind the carved screens. A porcelain cup clicks onto the stone tabletop, followed by the slight shuffle of boots on stone.\n\nnon_diegetic_music: A sustained low erhu drone at a slow tempo, joined by a single sparse pipa pluck that fades without resolution.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787665100787,
-    "promptEn": "constable politely asks the woman behind the screen for tea",
-    "lastPlayedAt": 1788331311063,
-    "playCount": 22
-  },
-  {
-    "id": "d2c77b87b84b",
-    "source": "h3console",
-    "prompt": "捕快隔着屏风跟女子说话",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe of layered sheer silk with a cream sash, with pale skin, a high black hair bun with loose strands framing her face, and calm unblinking eyes.\n<Subject 2> is the young male constable in <Picture 3>, wearing a dark indigo robe with a worn leather belt, a dao sword at his waist, a wooden warrant tag hanging at his hip, black boots and a black head wrap.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens, a single candle on a round marble-topped table, a wooden couch along the right wall, a hanging paper lantern, and deep shadows across the plaster walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> doing what <Subject 1> asked of him, lowering himself onto the stool beside the table, while keeping his weight forward, his hand near the dao at his waist, and his eyes tracking <Subject 1> the whole time. He complies without relaxing.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, candle on the marble table, hanging lantern, wooden couch, and ambient lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the white layered robe, cream sash, high black hair bun, facial features and pale complexion from <Picture 2> are retained throughout.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark indigo robe, leather belt, dao sword, warrant tag, black head wrap and facial features from <Picture 3> are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber candlelight against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens filling the left midground, a single candle burning on the round marble-topped table in the foreground, and the hanging paper lantern glowing softly against the cracked plaster wall on the right. The camera holds a static shot with small amplitude shake. <Subject 2>, the young constable, stands just inside frame right, his dark indigo robe still settling from movement. <Subject 1>, the woman in the white layered robe, stands near the screens on the left, one hand raised in a small, slow gesture toward the stool by the table, her expression composed and her gaze steady on him. <Subject 2> holds still for a beat, his jaw tight, then lowers his chin in a shallow nod. He steps forward and begins to sit on the low round stool beside the marble table, but he does not settle back. His weight stays forward on the balls of his feet, his left hand comes to rest on the scabbard of the dao at his waist, and his fingers close around it. As he sits, his head does not turn away from <Subject 1> for a moment. His eyes stay locked on her, following her as she lowers her raised hand and takes a single slow step back toward the shadow between two screen panels. The candle flame leans and stretches, throwing his elongated shadow across the floor and briefly swelling the darkness behind her. The camera pushes in with small amplitude at slow speed, tightening on <Subject 2> seated in the amber pool of candlelight, his shoulders squared, his hand still on the sword, while <Subject 1> stands pale and motionless in the half-dark beyond the carved wood, watching him back.\n\noverall_soundscape: The soft creak of wood as the constable lowers himself onto the stool, a faint leather scuff as his hand closes on the scabbard, the dry sputter of the candle wick, and the low settling creak of old floorboards. Two sets of breathing, one measured and one held.\n\nnon_diegetic_music: A sustained low erhu drone at a slow tempo, joined by a single sparse pipa pluck that fades without resolution.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787652321725,
-    "lastPlayedAt": 1788328362725,
-    "promptEn": "constable talking to the woman through the screen",
-    "playCount": 21
-  },
-  {
-    "id": "31ade4712105",
-    "source": "h3console",
-    "prompt": "术士念咒召出巨型蛤蟆，把白衣女鬼一口吞掉",
-    "fullPrompt": "subject_definitions: <Subject 1> is the female ghost in <Picture 1> — pale white flowing robes, long straight black hair, and a faint painted-mask texture of hairline cracks across her face, giving her an eerie, doll-like beauty. <Subject 2> is the giant summoned toad in <Picture 4.png>, Hama Wentai — an enormous golden-brown toad with warty skin, wearing a tattered Taoist robe and draped with talisman pouches and treasure charms, scaled large enough to burst through a room's double doors.\n\nsummary: [reference generation] The target video shows a summoner chanting an incantation to open a magic circle, from which the enormous toad Subject 2 bursts through a room's door with overwhelming presence, confronts the white-robed ghost Subject 1, delivers one decisive lunging strike, and swallows her whole to seal her.\n\nretention_analysis: <Subject 1> (appears in [Shot 3], [Shot 4]): fully_preserved - the white robes, long black hair, and cracked painted-mask facial texture are retained. <Subject 2> (appears in [Shot 1], [Shot 2], [Shot 3], [Shot 4]): fully_preserved - the golden-brown warty skin, tattered Taoist robe, and talisman pouches are retained.\n\ndetailed_description: Live-action, cinematic dark-fantasy Xianxia style, low-key lighting with cold blue ambient tones cut by warm golden flashes from the magic circle. [Shot 1] A hooded summoner (S1) kneels in a dim room, both palms pressed to the floor as a glowing golden magic circle blooms outward beneath them, talisman script igniting in sequence. The summoner (S1) says: <d>[Chinese] 山中蛤蟆，应劫而出，助我诛邪！</d> The camera pushes in with small amplitude at slow speed toward the brightening circle. [Shot 2] At 00:04.000, the shot cuts to the room's double doors, which splinter and burst outward as <Subject 2> forces its enormous bulk through the opening, wood fragments scattering, its Taoist robe swaying with the motion. The camera tilts up quickly to reveal the toad's full towering scale, ending on a low-angle framing. [Shot 3] At 00:08.000, the shot cuts to <Subject 1> standing rigid across the room, her expression shifting from composure to alarm as <Subject 2> plants its front legs and lunges directly at her in one continuous bounding strike. She lets out a sharp shriek. The camera does a fast Arc Shot with large amplitude, circling the collision point to emphasize the impact. [Shot 4] At 00:12.000, the shot cuts to <Subject 2> catching <Subject 1> mid-lunge and swallowing her whole in a single motion, her white robe vanishing past its jaws as talisman light flares briefly around its throat, sealing her. The camera holds a Static Shot on the toad settling back, robe and talismans still swaying, as the magic circle beneath it dims to a faint glow.\n\noverall_soundscape: Splintering wood as the door bursts open, heavy footfalls and displaced air as the toad lunges, a low guttural roar layered under the strike, the ghost's sharp shriek cut short mid-scream, and a crackling talisman-seal chime as she is swallowed, all against a faint ambient room echo throughout.\n\nnon_diegetic_music: A tense Xianxia orchestral score built on low guzheng plucks and taiko-style percussion, swelling sharply into a full brass-and-drum hit at the moment of the strike, then dropping to a single sustained low string tone as the seal completes.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787642579766,
-    "promptEn": "a summoner chants and calls a giant toad, which swallows the white-robed ghost whole",
-    "lastPlayedAt": 1788331008814,
-    "playCount": 22
-  },
-  {
-    "id": "22a666455198",
-    "source": "h3console",
-    "prompt": "我召唤蛤蟆文太打赢了画皮女鬼",
-    "fullPrompt": "我在这个游戏里召唤了一只仙家的蛤蟆 叫做蛤蟆文太，它跳出来跟角色 1 进行打斗，收服了角色 1 这个画皮女鬼，整个打斗场景出现在图片 2 的背景里",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787640567001,
-    "promptEn": "summoned toad Wentai, beat the painted-skin female ghost",
-    "lastPlayedAt": 1788330130098,
-    "playCount": 22
-  },
-  {
-    "id": "1cc42a8669a1",
-    "source": "h3console",
-    "prompt": "捕快隔着屏风跟女子说话",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman in <Picture 2>, wearing a flowing white traditional Chinese robe of layered sheer silk with a cream sash, with pale skin, a high black hair bun with loose strands framing her face, and calm unblinking eyes.\n<Subject 2> is the young male constable in <Picture 3>, wearing a dark indigo robe with a worn leather belt, a dao sword at his waist, a wooden warrant tag hanging at his hip, black boots and a black head wrap.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens, a single candle on a round marble-topped table, a wooden couch along the right wall, a hanging paper lantern, and deep shadows across the plaster walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> doing what <Subject 1> asked of him, lowering himself onto the stool beside the table, while keeping his weight forward, his hand near the dao at his waist, and his eyes tracking <Subject 1> the whole time. He complies without relaxing.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, candle on the marble table, hanging lantern, wooden couch, and ambient lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the white layered robe, cream sash, high black hair bun, facial features and pale complexion from <Picture 2> are retained throughout.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark indigo robe, leather belt, dao sword, warrant tag, black head wrap and facial features from <Picture 3> are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber candlelight against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens filling the left midground, a single candle burning on the round marble-topped table in the foreground, and the hanging paper lantern glowing softly against the cracked plaster wall on the right. The camera holds a static shot with small amplitude shake. <Subject 2>, the young constable, stands just inside frame right, his dark indigo robe still settling from movement. <Subject 1>, the woman in the white layered robe, stands near the screens on the left, one hand raised in a small, slow gesture toward the stool by the table, her expression composed and her gaze steady on him. <Subject 2> holds still for a beat, his jaw tight, then lowers his chin in a shallow nod. He steps forward and begins to sit on the low round stool beside the marble table, but he does not settle back. His weight stays forward on the balls of his feet, his left hand comes to rest on the scabbard of the dao at his waist, and his fingers close around it. As he sits, his head does not turn away from <Subject 1> for a moment. His eyes stay locked on her, following her as she lowers her raised hand and takes a single slow step back toward the shadow between two screen panels. The candle flame leans and stretches, throwing his elongated shadow across the floor and briefly swelling the darkness behind her. The camera pushes in with small amplitude at slow speed, tightening on <Subject 2> seated in the amber pool of candlelight, his shoulders squared, his hand still on the sword, while <Subject 1> stands pale and motionless in the half-dark beyond the carved wood, watching him back.\n\noverall_soundscape: The soft creak of wood as the constable lowers himself onto the stool, a faint leather scuff as his hand closes on the scabbard, the dry sputter of the candle wick, and the low settling creak of old floorboards. Two sets of breathing, one measured and one held.\n\nnon_diegetic_music: A sustained low erhu drone at a slow tempo, joined by a single sparse pipa pluck that fades without resolution.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787639957338,
-    "lastPlayedAt": 1788330846689,
-    "promptEn": "constable talking to the woman through the screen",
-    "playCount": 22
-  },
-  {
-    "id": "9f9638cb5b62",
-    "source": "h3console",
-    "prompt": "捕快慌张闯进屋，看到屏风后的女子",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 3> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman partially visible behind the carved wooden screens in <Picture 3>, wearing a flowing white traditional Chinese robe, with pale skin, long black hair, and an unnaturally still posture.\n<Subject 2> is a young male constable in a dark indigo robe with a dao sword at his waist, not present in <Picture 3>, who enters the room from outside.\n<Picture 3> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens, a single oil lamp on a low table, and deep shadows across the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> bursting through the doorway into the room established by <Picture 1>, stumbling forward in a panicked state, glancing back at the entrance, then turning to discover <Subject 1> standing motionless behind the screens. The oil lamp flame flickers violently from the sudden draft.\n\nretention_analysis:\n<Picture 3> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, oil lamp, wall textures, and ambient lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the white robe, long black hair, pale complexion, and still posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark indigo robe, dao sword, and constable appearance are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber lamp light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 3>, a medium-wide shot of the dim old room with five carved wooden screens in the midground, a single oil lamp casting flickering amber light on the low wooden table, and deep shadow pools along the cracked plaster walls. The camera holds a static shot with small amplitude shake. From the right edge of frame, <Subject 2>, the young constable in a dark indigo robe with a dao sword at his waist, stumbles through the wooden doorway. His right hand grips the door frame as his body lurches forward, his dark robe swinging with momentum. A gust of outside air causes the oil lamp flame to stretch sideways and throw long, rapidly shifting shadows across the screens. <Subject 2> steadies himself, breathing hard, and turns his head sharply back toward the open door, his jaw clenched and eyes wide. He holds that look for a beat, then slowly rotates forward. The camera pushes in with small amplitude at slow speed as his gaze lands on the carved screens. Through the gaps between screen panels, the faint silhouette of <Subject 1>, the woman in a white robe with long black hair, stands perfectly motionless. The oil lamp flame settles back to a gentle sway, casting a warm glow on <Subject 2>'s tense face while <Subject 1> remains a pale, still shape in the amber-lit darkness behind the carved wood.\n\noverall_soundscape: A heavy wooden door slams open followed by rapid boot steps on stone flooring. Wind gusts through the doorway and the oil lamp flame hisses. Heavy breathing gradually steadies as the room settles into quiet, leaving only the faint creak of old wood and the soft sputter of the lamp wick.\n\nnon_diegetic_music: A sustained low erhu drone at a slow tempo, joined by a single sparse pipa pluck that fades without resolution.",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787637292741,
-    "promptEn": "constable bursts in panicked, spots the woman behind the screen",
-    "lastPlayedAt": 1788329827838,
-    "playCount": 22
-  },
-  {
-    "id": "93c2b84a2a10",
-    "source": "h3console",
-    "prompt": "捕快慌张闯进屋，看到屏风后的女子",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 3> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman partially visible behind the carved wooden screens in <Picture 3>, wearing a flowing white traditional Chinese robe, with pale skin, long black hair, and an unnaturally still posture.\n<Subject 2> is a young male constable in a dark indigo robe with a dao sword at his waist, not present in <Picture 3>, who enters the room from outside.\n<Picture 3> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens, a single oil lamp on a low table, and deep shadows across the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> bursting through the doorway into the room established by <Picture 1>, stumbling forward in a panicked state, glancing back at the entrance, then turning to discover <Subject 1> standing motionless behind the screens. The oil lamp flame flickers violently from the sudden draft.\n\nretention_analysis:\n<Picture 3> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, oil lamp, wall textures, and ambient lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the white robe, long black hair, pale complexion, and still posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark indigo robe, dao sword, and constable appearance are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber lamp light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 3>, a medium-wide shot of the dim old room with five carved wooden screens in the midground, a single oil lamp casting flickering amber light on the low wooden table, and deep shadow pools along the cracked plaster walls. The camera holds a static shot with small amplitude shake. From the right edge of frame, <Subject 2>, the young constable in a dark indigo robe with a dao sword at his waist, stumbles through the wooden doorway. His right hand grips the door frame as his body lurches forward, his dark robe swinging with momentum. A gust of outside air causes the oil lamp flame to stretch sideways and throw long, rapidly shifting shadows across the screens. <Subject 2> steadies himself, breathing hard, and turns his head sharply back toward the open door, his jaw clenched and eyes wide. He holds that look for a beat, then slowly rotates forward. The camera pushes in with small amplitude at slow speed as his gaze lands on the carved screens. Through the gaps between screen panels, the faint silhouette of <Subject 1>, the woman in a white robe with long black hair, stands perfectly motionless. The oil lamp flame settles back to a gentle sway, casting a warm glow on <Subject 2>'s tense face while <Subject 1> remains a pale, still shape in the amber-lit darkness behind the carved wood.\n\noverall_soundscape: A heavy wooden door slams open followed by rapid boot steps on stone flooring. Wind gusts through the doorway and the oil lamp flame hisses. Heavy breathing gradually steadies as the room settles into quiet, leaving only the faint creak of old wood and the soft sputter of the lamp wick.\n\nnon_diegetic_music: A sustained low erhu drone at a slow tempo, joined by a single sparse pipa pluck that fades without resolution.",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787635935502,
-    "promptEn": "constable bursts in panicked, spots the woman behind the screen",
-    "lastPlayedAt": 1788330288233,
-    "playCount": 22
-  },
-  {
-    "id": "b32769bf80db",
-    "source": "h3console",
-    "prompt": "捕快慌张闯进屋，看到屏风后的女子",
-    "fullPrompt": "For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nsubject_definitions:\n<Subject 1> is the mysterious woman partially visible behind the carved wooden screens in <Picture 1>, wearing a flowing white traditional Chinese robe, with pale skin, long black hair, and an unnaturally still posture.\n<Subject 2> is a young male constable in a dark indigo robe with a dao sword at his waist, not present in <Picture 1>, who enters the room from outside.\n<Picture 1> is the first frame of [Shot 1], establishing the dim interior of the old room with five carved wooden screens, a single oil lamp on a low table, and deep shadows across the walls.\n\nsummary:\n[keyframe completion + reference generation] The target video shows <Subject 2> bursting through the doorway into the room established by <Picture 1>, stumbling forward in a panicked state, glancing back at the entrance, then turning to discover <Subject 1> standing motionless behind the screens. The oil lamp flame flickers violently from the sudden draft.\n\nretention_analysis:\n<Picture 1> ([Shot 1] first frame): fully_preserved - the room layout, carved screens, oil lamp, wall textures, and ambient lighting are retained as the spatial anchor.\n<Subject 1> (appears in [Shot 1]): fully_preserved - the white robe, long black hair, pale complexion, and still posture behind the screens are retained.\n<Subject 2> (appears in [Shot 1]): fully_preserved - the dark indigo robe, dao sword, and constable appearance are retained throughout.\n\ndetailed_description:\nThe target video is in a cinematic Chinese period-horror style with warm amber lamp light against deep black shadows and a slightly desaturated color palette.\n[Shot 1] The scene opens from the composition established by <Picture 1>, a medium-wide shot of the dim old room with five carved wooden screens in the midground, a single oil lamp casting flickering amber light on the low wooden table, and deep shadow pools along the cracked plaster walls. The camera holds a static shot with small amplitude shake. From the right edge of frame, <Subject 2>, the young constable in a dark indigo robe with a dao sword at his waist, stumbles through the wooden doorway. His right hand grips the door frame as his body lurches forward, his dark robe swinging with momentum. A gust of outside air causes the oil lamp flame to stretch sideways and throw long, rapidly shifting shadows across the screens. <Subject 2> steadies himself, breathing hard, and turns his head sharply back toward the open door, his jaw clenched and eyes wide. He holds that look for a beat, then slowly rotates forward. The camera pushes in with small amplitude at slow speed as his gaze lands on the carved screens. Through the gaps between screen panels, the faint silhouette of <Subject 1>, the woman in a white robe with long black hair, stands perfectly motionless. The oil lamp flame settles back to a gentle sway, casting a warm glow on <Subject 2>'s tense face while <Subject 1> remains a pale, still shape in the amber-lit darkness behind the carved wood.\n\noverall_soundscape: A heavy wooden door slams open followed by rapid boot steps on stone flooring. Wind gusts through the doorway and the oil lamp flame hisses. Heavy breathing gradually steadies as the room settles into quiet, leaving only the faint creak of old wood and the soft sputter of the lamp wick.\n\nnon_diegetic_music: A sustained low erhu drone at a slow tempo, joined by a single sparse pipa pluck that fades without resolution.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787632587351,
-    "promptEn": "constable bursts in panicked, spots the woman behind the screen",
-    "lastPlayedAt": 1788330804642,
-    "playCount": 22
-  },
-  {
-    "id": "282281777d1b",
-    "source": "h3console",
-    "prompt": "机械臂抓起红色方块放下",
-    "fullPrompt": "固定镜头，机器臂下降，抓起红色方块，向前移动一段距离后放下",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787615243753,
-    "promptEn": "robot arm picks up a red block and puts it down",
-    "lastPlayedAt": 1788330774622,
-    "playCount": 22
-  },
-  {
-    "id": "d040eec16a88",
-    "source": "h3console",
-    "prompt": "乙女游戏男主角展示PV",
-    "fullPrompt": "生成一支 乙游男主角色 PV。角色参考图：使用图2作为严格的角色身份参考。请保持同一个角色脸、发型、体型比例、服装设计、材质细节和乙游 CG 质感一致。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787615242602,
-    "promptEn": "otome game male lead showcase PV",
-    "lastPlayedAt": 1788330882706,
-    "playCount": 22
-  },
-  {
-    "id": "423a20afba5a",
-    "source": "h3console",
-    "prompt": "仙侠风角色转场，只露背影和侧脸",
-    "fullPrompt": "使用图2作为固定人物参考，保持黑色半扎长发、银色镂空发冠、黛蓝发带、浅色层叠汉服、半透明蓝色外袍、深蓝腰封、银色花形扣饰和长流苏一致。使用图1作为镜头分镜、节奏参考。画面为高质量 4K 16:9 国风 3D，电影级仙侠质感，热血、庄严、宿命感强。镜头依次表现分镜中的内容，确保每个分镜之间运镜、切换自然，不能像ppt一样。人物露脸只能是近景或特写，远景只能是背影、侧背或环境空镜，不要正脸远景。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787615241814,
-    "promptEn": "xianxia character turning, only back and side profile shown",
-    "lastPlayedAt": 1788326157348,
-    "playCount": 21
-  },
-  {
-    "id": "1a398591663e",
-    "source": "h3console",
-    "prompt": "黏土狐狸助跑冲下悬崖，慢动作飞跃峡谷",
-    "fullPrompt": "黏土动画风格。狂奔的狐狸冲到了悬崖边缘，毫不犹豫地腾空而起，在巨大的熔岩峡谷上方完成了一个极具张力的、慢动作的英雄式飞跃。镜头在半空中从狐狸的腹部下方极速掠过（大动态运镜），完美展现出深渊的恐怖深度，以及狐狸黏土身体在空中极致舒展的动态瞬间。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787615240988,
-    "promptEn": "clay fox sprints off a cliff, slow-mo leap over the canyon",
-    "lastPlayedAt": 1788330150115,
-    "playCount": 22
-  },
-  {
-    "id": "db125dc6a831",
-    "source": "h3console",
-    "prompt": "网站UI标题滑入，车灯由暗变红",
-    "fullPrompt": "模拟网站UI设计，先是上面的标题向下滑动出现，然后是下面的文字栏向上划出，汽车灯由暗变红",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787615240199,
-    "lastPlayedAt": 1788330066043,
-    "promptEn": "website UI title slides in, car light fades to red",
-    "playCount": 22
-  },
-  {
-    "id": "234a2037e064",
-    "source": "h3console",
-    "prompt": "产品官网页面炫酷下滑展示",
-    "fullPrompt": "一个网站页面，网站页面UI设计，网站动效，视频展示了流畅的网页向下滚动效果。一个极具爆发力与动感的产品官网风格产品落地页 UI/UX 演示视频，核心展示主体是该产品图片1。页面采用粗犷有力、倾斜的超大号无衬线字体进行张扬的排版。背景有极具速度感的动态光影、暗色碳纤维或运动透气网眼纹理在交织变换。视频展示了节奏紧凑、充满力量感的网页向下滚动效果，以及鼠标悬停时强烈的视觉放大与颜色反转等 UI 交互动作。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787588547172,
-    "promptEn": "product website scrolling down with flashy effects",
-    "lastPlayedAt": 1788330970782,
-    "playCount": 22
-  },
-  {
-    "id": "2e18bf09b46a",
-    "source": "h3console",
-    "prompt": "游戏菜单UI，角色换装机械手臂",
-    "fullPrompt": "角色参考图1，UI风格参考图2。[0 秒-2 秒] 高角度俯拍。角色坐在高饱和度的亮紫色地面上，参考为图1，抬头看向摄像机。右侧显示游戏菜单 UI：开始新游戏 (START NEW GAME)、继续游戏 (CONTINUE, 高亮显示)、设置 (SETTINGS)、退出游戏 (EXIT GAME)。左上角显示玩家资料 MINIMAX。光标点击“继续游戏”。[2 秒-4 秒] 平滑变焦至她的右臂。UI 面板从右侧滑入，“右臂装备 (RIGHT ARM EQUIPMENT)” 面板出现。选中高亮显示 “幻影之握 (PHANTOM GRIP)”，随后滑动至 “时空之爪 (CHRONOS CLAW)”。她的右手机械地重新配置，手指分开，新的爪状指节锁定到位，青色 LED 灯闪烁得更亮。[4 秒-7 秒] 摄像机平滑地绕到她的左侧。新的 UI 滑入 “武器定制 (ARMAMENT CUSTOMIZATION)” 网格，显示手部、前臂、肘部、上臂组件。选区快速循环切换部件。她的左臂逐段拆解：前臂板脱落，新装甲滑入，肘关节更换，手部重新配置，更换过程中可见暴露的线路和活塞。[7 秒-8.5 秒] 摄像机拉回至中景。确认配置 (CONFIRM CONFIG) 按钮闪烁。点击。所有 UI 面板向内收缩消失。角色双腿解开，调整姿势，现在放松地坐着，单膝立起，义肢手自然抬起并完成配置后的轻微动作展示。[8.5 秒-10 秒] 底部出现加载 (LOADING) 条。从 0% 快速填充至 100%。高饱和度的亮紫色环境开始变暗，阴影从边缘蔓延，温暖的金色光芒渗入。[10 秒-15 秒] 当她起身时，完整的环境在她周围加载。密集的赛博朋克贫民窟显现，霓虹灯闪烁，潮湿的街道反射着光芒，人群涌动，摩托车穿梭，错综复杂的电线，堆叠的建筑向远处的未来派摩天大楼延伸。摄像机稳定在第三人称视角，位于她身后，展示她的背部。HUD 元素淡入，右上角显示小地图，左下角显示生命值和弹药计数器。任务标记出现。她迈步走进街道。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787588545801,
-    "promptEn": "game menu UI, character swaps out a mechanical arm",
-    "lastPlayedAt": 1788326551672,
-    "playCount": 21
-  },
-  {
-    "id": "fe749aa533a0",
-    "source": "h3console",
-    "prompt": "人体工学椅360度展示广告",
-    "fullPrompt": "高级办公室Herman Miller黑色人体工学椅360度展示。透气网背特写空气流动可视化。腰部支撑+人体工学曲线工程动画。多维度调节扶手/座椅高度演示。设计师/程序员/创意人员长时间办公场景,姿态舒展自如。久坐不累3D骨骼支撑示意图。空间美学室内陈列。结尾“灵感与舒适同时在线”字样浮现。高端简约、冷色调、专业未来感、慢节奏运镜。产品详情参考图1，产品图参考图2。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787588545243,
-    "promptEn": "ergonomic office chair 360 showcase ad",
-    "lastPlayedAt": 1788327314329,
-    "playCount": 21
-  },
-  {
-    "id": "b0d258352c3c",
-    "source": "h3console",
-    "prompt": "白棚拍摄两位模特戴未来感眼镜",
-    "fullPrompt": "生成一支 竖屏 9:16 高级时尚眼镜广告片，整体参考所给视频的分镜节奏、剪辑速度、白棚质感和冷峻时尚氛围。画面是极简白棚，无缝白色背景，强烈的高定广告感，干净、简约、帅气、先锋、国际一线时装大片质感。主视觉人物参考图片1，两位全身女模特，一个黑人女模特，一个欧美女模特，保持她们的服装高级感、身体姿态、白棚光影、时装秀场气质和整体冷峻态度。两人都佩戴未来感高级眼镜，眼镜设计参考图3，强调包覆式弧面、锐利几何猫眼/护目镜混合轮廓、镜面反射、流线型镜腿、高级时尚配饰质感。两个人物的外貌细节参考图2。",
-    "edited": true,
-    "width": 720,
-    "height": 1280,
-    "duration": 12.5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787588544667,
-    "promptEn": "white studio shoot, two models wearing futuristic glasses",
-    "lastPlayedAt": 1788330172141,
-    "playCount": 22
-  },
-  {
-    "id": "71a4d093aebc",
-    "source": "h3console",
-    "prompt": "吸血鬼爱情短剧预告，女主误闯古堡",
-    "fullPrompt": "生成一支 15 秒、9:16 竖屏海外真人吸血鬼爱情短剧预告片段。男女主外形参考图1，场景参考图2。保持男女主身份一致，真人质感，高级短剧质感。整体故事：小白花人类女主误入古堡禁区，意外唤醒沉睡的吸血鬼贵族男主。男主发现她身上带着某种与古老战争有关的气息，因此对她产生强烈的控制欲和危险兴趣。女主害怕他，却没有彻底屈服，反抗他的压迫。整体风格：海外 ReelShort / DramaBox 吸血鬼爱情短剧预告感。暗黑浪漫、危险吸引力、宿命感、强控制感、阴郁压迫、高光反转。画面高级、克制、紧凑，像爆款短剧的前 15 秒钩子。不要血腥，不要廉价恐怖，不要万圣节感，不要现代街头感。画幅：9:16 竖屏构图，适合 TikTok / ReelShort / DramaBox。人物以中近景、近景、特写为主，竖屏里要突出脸、眼神、压迫感和关系张力。",
-    "edited": true,
-    "width": 720,
-    "height": 1280,
-    "duration": 12.5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787588544109,
-    "promptEn": "vampire romance drama trailer, girl wanders into the castle",
-    "lastPlayedAt": 1788329679709,
-    "playCount": 21
-  },
-  {
-    "id": "7ded7db676d3",
-    "source": "h3console",
-    "prompt": "MiniMax H3品牌科技感短片，蓝色波形logo",
-    "fullPrompt": "短片的准确开场画面。完整保留电光蓝背景、左侧白色“MiniMax”、中央浅蓝章鱼Logo和右侧白色“H3”。图片2:章鱼Logo完成线性形变后的目标波形Logo参考，也是后半段贯穿画面的核心图形。严格保持波形Logo的路径结构、弯曲数量、圆角端点和整体比例。生成一支严格15秒、16:9、4K、60fps的MiniMax H3抽象科技MG品牌短片。短片表达H3从专用任务走向通用多模态模型的设计理念：语言、图片、视频和音频不再彼此隔离，而是被同一个上下文、同一套架构和同一根连续线连接起来。整体风格高级、简洁、灵动、有艺术感。采用大面积电光蓝、深黑蓝、浅青色和少量紫色色块，搭配干净的白色粗体文字。画面具有大量负空间，避免复杂图标和信息堆叠。全片的核心运动元素是一根浅蓝色连续线，以及位于线条前端的白色发光节点。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787588543551,
-    "promptEn": "MiniMax H3 tech brand film, blue waveform logo",
-    "lastPlayedAt": 1788330870695,
-    "playCount": 22
-  },
-  {
-    "id": "81711242178d",
-    "source": "h3console",
-    "prompt": "悬疑犯罪片头，拼贴分屏字幕",
-    "fullPrompt": "生成一支15秒16:9横版轻悬疑犯罪电影片头。整体风格参考这些图的视觉语言：复古日系动画片头、硬边剪影、漫画拼贴、非对称分屏、强烈几何色块、英文片头署名、少量日文片假名装饰、爵士犯罪气质。氛围是60%悬疑、40%爵士：神秘、酷、灵动、有都市犯罪感，不恐怖、不沉重，也不要变成欢乐爵士MV。片头动效要像动态图形拼贴：黑底线框先出现，分屏边界快速划出，色块和画格逐块贴入；人物剪影、道具特写、英文署名跟随鼓点依次滑入、弹出、遮罩揭示。不要普通写实剧情动画，要片头包装感。英文署名要清晰可读，允许有动效：细线框先画出，名字从框内滑入，字母逐个出现，被色块遮罩揭示，最后短暂停住。不要新增中文，不要乱码，不要让英文错拼。全片规则：所有英文署名和职位只出现一次。不要重复同一个职位，不要重复同一个人名，不要让同一个人担任多个职位。转场形式要丰富：圆形黑胶遮罩、车门竖切、人物长影擦屏、红线切割、巨大英文字母遮罩、分屏边框重组、色块硬切、画格逐块贴入。所有转场跟随鼓点，干脆、悬疑、灵动、有漫画拼贴感。不要柔和溶解，不要流体转场。BGM：原创15秒片头音乐，60%悬疑、40%爵士。以低音持续音、紧张弦乐拨奏、冷感合成器脉冲、低音鼓、稀疏爵士鼓刷、walking bass片段、低音萨克斯短句和短促铜管点缀组成。前2秒神秘低频和踩镲建立悬疑，第3秒进入低沉鼓点，第6秒加入爵士贝斯律动，第10秒出现短促萨克斯/铜管riff，最后2秒用紧张和弦加鼓点定格。整体神秘、酷、带犯罪感，爵士只占40%，不要太欢快，不模仿任何现有旋律。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787588542968,
-    "promptEn": "mystery crime title sequence, collage split-screen credits",
-    "lastPlayedAt": 1788327740651,
-    "playCount": 21
-  },
-  {
-    "id": "67ababbea78f",
-    "source": "h3console",
-    "prompt": "时尚男子墨镜风衣站在火场旁",
-    "fullPrompt": "图1是整体质感、氛围参考，图2是人物外表特征参考。生成一支 15 秒 16:9 横版潮流时装短片。保持人物一致：白金色长发、窄黑色复古墨镜、黑色高光漆皮风衣、冷酷自信的时尚表情、火光映在皮衣表面的橙色反射。整体风格是胶片快切时尚广告，融合夜晚火场、黑烟、橙红火焰、VHS 故障、CCTV 广播中断、90 年代模拟胶片颗粒、扫描线、色差、漏光、闪白转场、画面轻微抖动。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787588542426,
-    "promptEn": "fashion guy in sunglasses and coat by a fire",
-    "lastPlayedAt": 1788331461191,
-    "playCount": 22
-  },
-  {
-    "id": "c1237b599bf4",
-    "source": "h3console",
-    "prompt": "女人在沙漠公路边打开后备箱拿包",
-    "fullPrompt": "制作一支 16:9 横版高级时尚品牌视频。整体氛围、场景和胶片质感参考图1；包袋资产参考图3；人物资产参考图2；品牌 ending logo 参考图4。这是一支卖衣服和包的时尚 campaign，整体气质要高级、冷感、克制，但剪辑不能无聊，要更灵动、更有时尚节奏。不要像普通剧情片，也不要像电商广告。核心故事依然很简单：荒漠公路和复古车旁，女人回到车尾，打开后备箱，取出黑色包袋，和站在车边的男人有一瞬间的安静关系，然后她拿着包离开。包和服装要自然融入人物行动里，成为角色气质的一部分。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787588541726,
-    "promptEn": "woman opens the trunk on a desert road, takes out a bag",
-    "lastPlayedAt": 1788331172948,
-    "playCount": 22
-  },
-  {
-    "id": "97846542ab09",
-    "source": "h3console",
-    "prompt": "巨大宇宙之门前站着小小背影，标题浮现",
-    "fullPrompt": "写实电影感，高反差光影，节奏紧凑。图1作为整体氛围、风格参考。图2作为主角人物参考。Shot 1 — 超广角建立镜头。巨大圆形宇宙门几乎占满画面，人物只是门前一个很小的背影，站在画面下方偏右。地面潮湿反光，门中心一片黑暗。镜头慢慢向前推进。大标题从黑暗边缘渐入，先模糊后清晰：“THE STARS WERE LISTENING” 字体极窄、厚重、全大写，暗红混合铁锈红，带轻微颗粒和雾化边缘。Audio: 深低频脉冲，远处金属轻震，文字清晰时一记轻 hit。→ Hard cut。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787588541164,
-    "promptEn": "tiny silhouette before a huge cosmic gate, title fades in",
-    "lastPlayedAt": 1788327132179,
-    "playCount": 21
-  },
-  {
-    "id": "5cc09dcdf470",
-    "source": "h3console",
-    "prompt": "望远镜视角搜寻MiniMax装置，时尚路人走过",
-    "fullPrompt": "图1至图4连续关键帧，模拟老式望远镜寻MINIMAX装置。开头虚焦手持晃动后快速拉近rack focus至图1，图间望远镜扫视转场（甩动、运动模糊、光学拖影、曝光闪烁）在最模糊处切换并迅速稳定重新对焦。全程固定双圆镜片遮罩（黑色羽化暗角、位置大小边缘虚化绝对一致无变形漂移）仅内部画面运动。图2布料随风克制飘动MINIMAX文字随布面起伏保持可读；图3人物如无意捕捉的时尚路人自然走动转身摆臂；图4人物调整眼镜或微抬下巴展现冷感松弛fashion campaign态度。红字随对焦轻微加载，初现略虚焦低透明度0.3–0.5秒内淡入变清晰（可微小纵向滑入或字距呼吸）切换前淡出或被运动模糊带走，禁旋转弹跳或大幅飞入飞出。Wes Anderson式窥视35mm胶片质感（轻微颗粒、柔和高光晕染、克制色彩、红字点缀），极简高级微带搞怪灵动，严禁新增人物车辆建筑logo，保持原图核心构图与MINIMAX装置不变。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787588540642,
-    "promptEn": "telescope view searching for a MiniMax device, fashionable passerby",
-    "lastPlayedAt": 1788327242275,
-    "playCount": 21
-  },
-  {
-    "id": "a59cae01ec39",
-    "source": "h3console",
-    "prompt": "望远镜视角搜寻MiniMax装置",
-    "fullPrompt": "图1至图4连续关键帧，模拟老式望远镜寻MINIMAX装置。开头虚焦手持晃动后快速拉近rack focus至图1，图间望远镜扫视转场（甩动、运动模糊、光学拖影、曝光闪烁）在最模糊处切换并迅速稳定重新对焦。全程固定双圆镜片遮罩（黑色羽化暗角、位置大小边缘虚化绝对一致无变形漂移）仅内部画面运动。图2布料随风克制飘动MINIMAX文字随布面起伏保持可读；图3人物如无意捕捉的时尚路人自然走动转身摆臂；图4人物调整眼镜或微抬下巴展现冷感松弛fashion campaign态度。红字随对焦轻微加载，初现略虚焦低透明度0.3–0.5秒内淡入变清晰（可微小纵向滑入或字距呼吸）切换前淡出或被运动模糊带走，禁旋转弹跳或大幅飞入飞出。Wes Anderson式窥视35mm胶片质感（轻微颗粒、柔和高光晕染、克制色彩、红字点缀），极简高级微带搞怪灵动，严禁新增人物车辆建筑logo，保持原图核心构图与MINIMAX装置不变。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787588540567,
-    "promptEn": "telescope view searching for a MiniMax device",
-    "lastPlayedAt": 1788326051252,
-    "playCount": 21
-  },
-  {
-    "id": "ba8c0758e5b4",
-    "source": "h3console",
-    "prompt": "地下说唱兄弟对着镜头freestyle",
-    "fullPrompt": "【参考素材说明】@图片3：场景视觉风格（街道/夜间/地下空间、压迫感构图、环境层次、影像颗粒）@图片2：文字包装样式（字体质感、图形设计、动态图形排版冲击力）@图片1：人物形象（脸、发型、服装轮廓、比例、气质、整体氛围）只参考指定维度，不直接复制参考图，不出现真实品牌、原参考图 logo/标题/可识别文字。\n\n【核心创意】10秒，16:9 横版 trap MV。两位 fly detective 兄弟搭挡在多个近景地下空间轮流对镜头 rap，全程跟随 trap 鼓点律动、卡点硬切，高反差印刷海报质感的英文块字随 bass hit 压屏出现。地下音乐录像带 + 时尚杂志拼贴 + 高时装质感，兄弟搭档式冷峻 performance。歌词围绕 VOKO 的核心概念展开：进入属于你的故事世界，做出选择，让世界回应你，最终落到 “YOUR KOSMO IS CALLING”。\n\n【画面过程描述】Shot 1 — 面部极近特写 / 通道压迫感- 场景：狭窄通道或地下入口近景，参考 @图片3- 景别：面部极近特写（脸/眼神/肩颈/衣领局部）- 人物：Detective A 直视镜头开始 rap，眼神冷静锐利，歌词：\"HEAR THAT? YOUR KOSMO CALLIN'.\"- 文字：\"KOSMO CALLIN'\" 巨大粗体英文压入画面上下，不遮挡眼睛- 律动：808 bass 砸下，\"KOSMO CALLIN'\" 瞬间纵向压缩后回弹；hi-hat roll 时字母边缘高速细碎震动；\"KOSMO\" 重音处扫描错位- 硬切Shot 2 — 中近景半身 / 墙面文字背景- 场景：另一处墙面/海报墙近景（和 Shot 1 空间明显不同）- 景别：明显拉到中近景半身- 人物：Detective B 对镜头 rap，歌词：\"DON'T JUST WATCH — WALK IN.\"，肩膀和头部跟 hi-hat 点拍，身体轻微前倾- 文字：\"WALK IN\" 巨大压缩英文在人物身后，被头发/肩膀/服装轮廓自然遮挡- 律动：文字纵向拉伸如海报立起；每个 snare 时文字突然放大、抖动、扫描错位- 硬切Shot 3 — 手部特写 / 遮挡切换- 场景：金属门边/栏杆/车库墙面/低矮天花局部- 景别：手部近景（手/戒指/袖口/衣服材质为主体）- 人物：Detective A 手势向镜头推进，像邀请观众进入，脸在后方虚化或局部，歌词：\"MAKE A MOVE — WORLD MOVES.\"- 文字：竖排 \"MAKE A MOVE\" 在前景一侧，被手部短暂遮挡- 律动：\"WORLD MOVES\" 时文字被拉成一条长竖条，跟 bass hit 硬切重组- 硬切Shot 4 — 肩颈特写到面部特写 / 楼梯转角- 场景：楼梯转角/门口阴影/低矮通道- 景别：肩颈和衣领材质特写 → snare 时突然短促推近到面部特写- 人物：Detective B 侧头靠近镜头继续 rap，歌词：\"NO SCRIPT — YOU CHOOSE.\"- 文字：\"YOU CHOOSE\" 像杂志标题被声音冲出来，在鼓点上突然放大- 律动：推近时文字位置硬切重排，始终不遮挡眼睛- 硬切Shot 5 — 半身近景残影 / 车库或混凝土背景- 场景：车库感空间或混凝土背景- 景别：Detective A 半身近景- 人物：主图层嘴型清楚同步，副图层动感残影（复印机错位/帧延迟，脸不变形），手势横向划过镜头，歌词：\"ONE CHOICE — NEW KOSMO.\"- 文字：竖排 \"NEW KOSMO\" 在背景跳帧闪现- 律动：\"CHOICE\" 瞬间横向切开画面触发闪白硬切；\"KOSMO\" 压缩成粗重块字后突然回弹- 硬切Shot 6 — 双人近景分屏 / 色块分割- 场景：地下入口/墙面/栏杆/门框（和 Shot 5 空间明显不同）- 景别：色块切成两/三区，两个 detective 面部特写和半身局部快速硬切交替- 人物：两 detective 在左右轮流表演，Detective A rap：\"YOUR STORY.\"，Detective B rap：\"YOUR WORLD.\"，两人共同：\"YOUR KOSMO.\"- 文字：中央出现 \"YOUR KOSMO\" 粗体短词- 律动：随 808 bass hit 硬切重组，字母先压扁再瞬间拉伸；hi-hat roll 时文字碎片快速跳动；最后重音上出现扫描错位和纸张撕裂感震动- 硬切Final — 多场景近景 performance montage- 场景：狭窄通道面部特写、墙面中近景、手部特写、楼梯转角肩颈特写、车库半身局部、门口阴影双人近景快速交替- 景别：禁止全身、禁止多人全景，只用近景/中近景/面部特写/手部特写/肩颈特写/半身镜头- 人物：两位 detective 轮流对镜头 rap，嘴型/下颌/呼吸/眉眼/手势都精准卡住 vocal、snare、hi-hat roll、bass hit。最后音乐短暂抽空，Detective A 贴近镜头低声 rap：\"YOU HEAR IT?\"，最后重拍两人共同 rap：\"YOUR KOSMO IS CALLING.\"- 文字：最终主文字 \"YOUR KOSMO IS CALLING\" 巨大压入画面- 律动：每次 bass hit 触发场景硬切或景别硬切；每次 snare 触发文字压屏或图层错位；最后一拍 \"YOUR KOSMO IS CALLING\" 随重音压屏，所有场景碎片同步冻结，hard cut to black，黑屏短暂出现 \"VOKO.WORLD\"\n\n【整体要求补充】\n▍影像风格（贯穿全片）- 粗颗粒、轻微胶片抖动、复印纸颗粒、半色调网点、印刷毛边、扫描错位、跳帧残影、高速图层错位- 剪辑节奏极快，只用硬切、跳切、鼓点切、遮挡切、闪白硬切、文字压屏硬切- 不用淡入淡出、不用柔和转场\n▍文字包装（贯穿全片）- 文字是画面主体的一部分，有真实图层关系- 文字可出现在人物前景/中景/背景之间- 人物会遮挡文字，文字也可压过人物局部，但不能遮挡眼睛和主要面部表情- 文字跟随 vocal 重音、snare、hi-hat roll、808 bass hit 出现/震动/拉伸/压缩/错位/撕裂/硬切重组- 风格：粗重压缩无衬线、巨大英文块字、竖向拉伸、横向压缩、纵排英文、弧形标题、黑白红高反差、复印纸颗粒、半色调网点、印刷破损、扫描错位、zine 拼贴感- 文字数量克制：每镜头只 1 个主文字元素，最多少量编号；不要密集小字、不要 logo\n▍律动规则（贯穿全片）- hi-hat roll → 快速微震、跳帧、文字细碎重组- snare → 文字猛然放大、画面硬切、人物肩膀下压- 808 bass hit → 低频压屏、画面短暂变形、文字纵向拉伸或横向压缩- vocal 关键词 → 嘴型同步、下颌动作、头部点拍、手势推进- 人物动作、文字动效、场景切换、景别切换必须共同卡点\n▍场景切换（贯穿全片）- 多个近景场景快速硬切- 场景气质统一参考 @图片3，但每个镜头空间要明显不同：狭窄通道、墙面近景、门口阴影、金属/混凝土背景、灯光边缘、海报墙、楼梯转角、低矮天花、车库感空间、地下入口- 不做大远景，不做复杂大场面- 每次切换由 bass hit、snare、vocal 重音或文字压屏触发\n▍景别切换（贯穿全片）- 变化要非常明显：面部极近特写 → 中近景半身 → 手部特写 → 肩颈特写 → 双人近景分屏 → 面部特写 → 半身局部- 不连续使用同一种景别- 每个硬切都让镜头距离/人物朝向/背景空间/文字层级产生明显变化- 镜头可轻微手持、突然推近、短促横移、快速压缩变焦、近距离摇摆- 不变成柔和慢镜头\n▍人物规则（贯穿全片）- 角色：两位 fly detectives / brothers- 保持 @图片1 人物形象的脸、发型、服装轮廓、比例、气质、整体氛围- 真实哑光皮肤、干净高级、自然毛孔、轻微皮肤纹理- 不要油亮 AI 美颜脸- 兄弟搭档式轮流 rap，对镜头有强烈 performance\n▍人物表演（贯穿全片）- 声音冷酷、攻击性、松弛但有压迫感，节奏紧、断裂、强拍- 不是静态摆拍，是在表演- 清楚嘴型同步：唇形、下颌、表情、呼吸跟随 vocal- 卡点动作：头部点拍、肩膀下压、手势推进、身体前倾、侧头、转身靠近镜头、手指指向镜头\n▍重要限制（贯穿全片）- 不用全身镜头- 不用多人全景- 不用复杂群像大场面- 全片只用近景/中近景/面部特写/手部特写/肩颈特写/半身镜头- 只表现人物上半身、脸部、手势、衣服材质、表情、嘴型同步、文字包装- 不要求看到完整身体和脚- 不出现真实品牌、原参考图中的 logo、标题或可识别原文字",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787588114381,
-    "promptEn": "underground rap duo freestyling at the camera",
-    "lastPlayedAt": 1788331030834,
-    "playCount": 22
-  },
-  {
-    "id": "58a439a93c02",
-    "source": "h3console",
-    "prompt": "男子闯进古宅道歉，屏风后有白衣女子",
-    "fullPrompt": "中式古风恐怖场景，昏暗的旧屋内，油灯微弱的光照亮一间陈旧的房间，屏风后隐约可见白衣女子的身影。一位身穿古装的男子对不起，我不是有意闯进来的，外面有东西在追我。。画面氛围诡异，光影交错，充满悬疑感。电影质感，4K画质。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787577878370,
-    "promptEn": "man barges into the old house apologizing, woman in white behind the screen",
-    "lastPlayedAt": 1788330694545,
-    "playCount": 22
-  },
-  {
-    "id": "0e75ef71c3d0",
-    "source": "h3console",
-    "prompt": "男子谢过女子收留，屏风后白衣女子",
-    "fullPrompt": "中式古风恐怖场景，昏暗的旧屋内，油灯微弱的光照亮一间陈旧的房间，屏风后隐约可见白衣女子的身影。一位身穿古装的男子我明白，多谢你收留我。。画面氛围诡异，光影交错，充满悬疑感。电影质感，4K画质。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787577088842,
-    "promptEn": "man thanks the woman for taking him in, behind the screen",
-    "lastPlayedAt": 1788327682606,
-    "playCount": 21
-  },
-  {
-    "id": "e656777b6eef",
-    "source": "h3console",
-    "prompt": "男子谢过女子收留自己",
-    "fullPrompt": "中式古风恐怖场景，昏暗的旧屋内，油灯微弱的光照亮一间陈旧的房间，屏风后隐约可见白衣女子的身影。一位身穿古装的男子我明白，多谢你收留我。。画面氛围诡异，光影交错，充满悬疑感。电影质感，4K画质。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787577088778,
-    "promptEn": "man thanks the woman for taking him in",
-    "lastPlayedAt": 1788326435580,
-    "playCount": 21
-  },
-  {
-    "id": "8d0270466c35",
-    "source": "h3console",
-    "prompt": "男子照她的话压低声音，观察门窗和油灯",
-    "fullPrompt": "中式古风恐怖场景，昏暗的旧屋内，油灯微弱的光照亮一间陈旧的房间，屏风后隐约可见白衣女子的身影。一位身穿古装的男子照她的话压低声音，先观察门窗、油灯和她的位置。。画面氛围诡异，光影交错，充满悬疑感。电影质感，4K画质。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787576943673,
-    "promptEn": "man lowers his voice like she said, checks the door, window and lamp",
-    "lastPlayedAt": 1788330424346,
-    "playCount": 22
-  },
-  {
-    "id": "396de15c1439",
-    "source": "h3console",
-    "prompt": "男子照她的话压低声音，观察门窗和油灯",
-    "fullPrompt": "中式古风恐怖场景，昏暗的旧屋内，油灯微弱的光照亮一间陈旧的房间，屏风后隐约可见白衣女子的身影。一位身穿古装的男子照她的话压低声音，先观察门窗、油灯和她的位置。。画面氛围诡异，光影交错，充满悬疑感。电影质感，4K画质。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787576943608,
-    "promptEn": "man lowers his voice like she said, checks the door, window and lamp",
-    "lastPlayedAt": 1788327352362,
-    "playCount": 21
-  },
-  {
-    "id": "cde82791f726",
-    "source": "h3console",
-    "prompt": "男子说愿意把心交给她",
-    "fullPrompt": "中式古风恐怖场景，昏暗的旧屋内，油灯微弱的光照亮一间陈旧的房间，屏风后隐约可见白衣女子的身影。一位身穿古装的男子我愿意把心交给你。。画面氛围诡异，光影交错，充满悬疑感。电影质感，4K画质。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787576798050,
-    "promptEn": "man says he's willing to give her his heart",
-    "lastPlayedAt": 1788330360291,
-    "playCount": 22
-  },
-  {
-    "id": "57f2c6fa351d",
-    "source": "h3console",
-    "prompt": "男子沉默地看着她，没接话",
-    "fullPrompt": "中式古风恐怖场景，昏暗的旧屋内，油灯微弱的光照亮一间陈旧的房间，屏风后隐约可见白衣女子的身影。一位身穿古装的男子沉默着看着她，没有立刻接话。。画面氛围诡异，光影交错，充满悬疑感。电影质感，4K画质。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787576651668,
-    "promptEn": "man silently looks at her, doesn't respond",
-    "lastPlayedAt": 1788331275028,
-    "playCount": 22
-  },
-  {
-    "id": "5958b86ab8f4",
-    "source": "h3console",
-    "prompt": "男子照她的话做，但保持警惕",
-    "fullPrompt": "中式古风恐怖场景，昏暗的旧屋内，油灯微弱的光照亮一间陈旧的房间，屏风后隐约可见白衣女子的身影。一位身穿古装的男子按对方的要求行动，但保持警惕。。画面氛围诡异，光影交错，充满悬疑感。电影质感，4K画质。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787576506447,
-    "lastPlayedAt": 1788329651687,
-    "promptEn": "man does as she says but stays on guard",
-    "playCount": 21
-  },
-  {
-    "id": "dd69b54fb2ce",
-    "source": "h3console",
-    "prompt": "男子道歉说自己被外面的东西追",
-    "fullPrompt": "中式古风恐怖场景，昏暗的旧屋内，油灯微弱的光照亮一间陈旧的房间，屏风后隐约可见白衣女子的身影。一位身穿古装的男子对不起，我不是有意闯进来的，外面有东西在追我。。画面氛围诡异，光影交错，充满悬疑感。电影质感，4K画质。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787576358982,
-    "lastPlayedAt": 1788330442352,
-    "promptEn": "man apologizes, says something outside was chasing him",
-    "playCount": 22
-  },
-  {
-    "id": "9fca8054035f",
-    "source": "h3console",
-    "prompt": "男子道歉说自己被外面的东西追",
-    "fullPrompt": "中式古风恐怖场景，昏暗的旧屋内，油灯微弱的光照亮一间陈旧的房间，屏风后隐约可见白衣女子的身影。一位身穿古装的男子对不起，我不是有意闯进来的，外面有东西在追我。。画面氛围诡异，光影交错，充满悬疑感。电影质感，4K画质。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787576013870,
-    "promptEn": "man apologizes, says something outside was chasing him",
-    "lastPlayedAt": 1788330678529,
-    "playCount": 22
-  },
-  {
-    "id": "7d425d2d35fb",
-    "source": "h3console",
-    "prompt": "男子道歉说自己被外面的东西追",
-    "fullPrompt": "中式古风恐怖场景，昏暗的旧屋内，油灯微弱的光照亮一间陈旧的房间，屏风后隐约可见白衣女子的身影。一位身穿古装的男子对不起，我不是有意闯进来的，外面有东西在追我。。画面氛围诡异，光影交错，充满悬疑感。电影质感，4K画质。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787575790848,
-    "promptEn": "man apologizes, says something outside was chasing him",
-    "lastPlayedAt": 1788331048845,
-    "playCount": 22
-  },
-  {
-    "id": "5e6dcf2c8cfd",
-    "source": "h3console",
-    "prompt": "男子道歉说自己被外面的东西追",
-    "fullPrompt": "中式古风恐怖场景，昏暗的旧屋内，油灯微弱的光照亮一间陈旧的房间，屏风后隐约可见白衣女子的身影。一位身穿古装的男子对不起，我不是有意闯进来的，外面有东西在追我。。画面氛围诡异，光影交错，充满悬疑感。电影质感，4K画质。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787575164928,
-    "promptEn": "man apologizes, says something outside was chasing him",
-    "lastPlayedAt": 1788331004808,
-    "playCount": 22
-  },
-  {
-    "id": "250fad90393c",
-    "source": "h3console",
-    "prompt": "男子道歉说自己被外面的东西追",
-    "fullPrompt": "中式古风恐怖场景，昏暗的旧屋内，油灯微弱的光照亮一间陈旧的房间，屏风后隐约可见白衣女子的身影。一位身穿古装的男子对不起，我不是有意闯进来的，外面有东西在追我。。画面氛围诡异，光影交错，充满悬疑感。电影质感，4K画质。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787575164766,
-    "promptEn": "man apologizes, says something outside was chasing him",
-    "lastPlayedAt": 1788326411562,
-    "playCount": 21
-  },
-  {
-    "id": "fb3000701384",
-    "source": "h3console",
-    "prompt": "地下说唱兄弟对着镜头freestyle",
-    "fullPrompt": "【参考素材说明】@图片3：场景视觉风格（街道/夜间/地下空间、压迫感构图、环境层次、影像颗粒）@图片2：文字包装样式（字体质感、图形设计、动态图形排版冲击力）@图片1：人物形象（脸、发型、服装轮廓、比例、气质、整体氛围）只参考指定维度，不直接复制参考图，不出现真实品牌、原参考图 logo/标题/可识别文字。\n\n【核心创意】10秒，16:9 横版 trap MV。两位 fly detective 兄弟搭挡在多个近景地下空间轮流对镜头 rap，全程跟随 trap 鼓点律动、卡点硬切，高反差印刷海报质感的英文块字随 bass hit 压屏出现。地下音乐录像带 + 时尚杂志拼贴 + 高时装质感，兄弟搭档式冷峻 performance。歌词围绕 VOKO 的核心概念展开：进入属于你的故事世界，做出选择，让世界回应你，最终落到 “YOUR KOSMO IS CALLING”。\n\n【画面过程描述】Shot 1 — 面部极近特写 / 通道压迫感- 场景：狭窄通道或地下入口近景，参考 @图片3- 景别：面部极近特写（脸/眼神/肩颈/衣领局部）- 人物：Detective A 直视镜头开始 rap，眼神冷静锐利，歌词：\"HEAR THAT? YOUR KOSMO CALLIN'.\"- 文字：\"KOSMO CALLIN'\" 巨大粗体英文压入画面上下，不遮挡眼睛- 律动：808 bass 砸下，\"KOSMO CALLIN'\" 瞬间纵向压缩后回弹；hi-hat roll 时字母边缘高速细碎震动；\"KOSMO\" 重音处扫描错位- 硬切Shot 2 — 中近景半身 / 墙面文字背景- 场景：另一处墙面/海报墙近景（和 Shot 1 空间明显不同）- 景别：明显拉到中近景半身- 人物：Detective B 对镜头 rap，歌词：\"DON'T JUST WATCH — WALK IN.\"，肩膀和头部跟 hi-hat 点拍，身体轻微前倾- 文字：\"WALK IN\" 巨大压缩英文在人物身后，被头发/肩膀/服装轮廓自然遮挡- 律动：文字纵向拉伸如海报立起；每个 snare 时文字突然放大、抖动、扫描错位- 硬切Shot 3 — 手部特写 / 遮挡切换- 场景：金属门边/栏杆/车库墙面/低矮天花局部- 景别：手部近景（手/戒指/袖口/衣服材质为主体）- 人物：Detective A 手势向镜头推进，像邀请观众进入，脸在后方虚化或局部，歌词：\"MAKE A MOVE — WORLD MOVES.\"- 文字：竖排 \"MAKE A MOVE\" 在前景一侧，被手部短暂遮挡- 律动：\"WORLD MOVES\" 时文字被拉成一条长竖条，跟 bass hit 硬切重组- 硬切Shot 4 — 肩颈特写到面部特写 / 楼梯转角- 场景：楼梯转角/门口阴影/低矮通道- 景别：肩颈和衣领材质特写 → snare 时突然短促推近到面部特写- 人物：Detective B 侧头靠近镜头继续 rap，歌词：\"NO SCRIPT — YOU CHOOSE.\"- 文字：\"YOU CHOOSE\" 像杂志标题被声音冲出来，在鼓点上突然放大- 律动：推近时文字位置硬切重排，始终不遮挡眼睛- 硬切Shot 5 — 半身近景残影 / 车库或混凝土背景- 场景：车库感空间或混凝土背景- 景别：Detective A 半身近景- 人物：主图层嘴型清楚同步，副图层动感残影（复印机错位/帧延迟，脸不变形），手势横向划过镜头，歌词：\"ONE CHOICE — NEW KOSMO.\"- 文字：竖排 \"NEW KOSMO\" 在背景跳帧闪现- 律动：\"CHOICE\" 瞬间横向切开画面触发闪白硬切；\"KOSMO\" 压缩成粗重块字后突然回弹- 硬切Shot 6 — 双人近景分屏 / 色块分割- 场景：地下入口/墙面/栏杆/门框（和 Shot 5 空间明显不同）- 景别：色块切成两/三区，两个 detective 面部特写和半身局部快速硬切交替- 人物：两 detective 在左右轮流表演，Detective A rap：\"YOUR STORY.\"，Detective B rap：\"YOUR WORLD.\"，两人共同：\"YOUR KOSMO.\"- 文字：中央出现 \"YOUR KOSMO\" 粗体短词- 律动：随 808 bass hit 硬切重组，字母先压扁再瞬间拉伸；hi-hat roll 时文字碎片快速跳动；最后重音上出现扫描错位和纸张撕裂感震动- 硬切Final — 多场景近景 performance montage- 场景：狭窄通道面部特写、墙面中近景、手部特写、楼梯转角肩颈特写、车库半身局部、门口阴影双人近景快速交替- 景别：禁止全身、禁止多人全景，只用近景/中近景/面部特写/手部特写/肩颈特写/半身镜头- 人物：两位 detective 轮流对镜头 rap，嘴型/下颌/呼吸/眉眼/手势都精准卡住 vocal、snare、hi-hat roll、bass hit。最后音乐短暂抽空，Detective A 贴近镜头低声 rap：\"YOU HEAR IT?\"，最后重拍两人共同 rap：\"YOUR KOSMO IS CALLING.\"- 文字：最终主文字 \"YOUR KOSMO IS CALLING\" 巨大压入画面- 律动：每次 bass hit 触发场景硬切或景别硬切；每次 snare 触发文字压屏或图层错位；最后一拍 \"YOUR KOSMO IS CALLING\" 随重音压屏，所有场景碎片同步冻结，hard cut to black，黑屏短暂出现 \"VOKO.WORLD\"\n\n【整体要求补充】\n▍影像风格（贯穿全片）- 粗颗粒、轻微胶片抖动、复印纸颗粒、半色调网点、印刷毛边、扫描错位、跳帧残影、高速图层错位- 剪辑节奏极快，只用硬切、跳切、鼓点切、遮挡切、闪白硬切、文字压屏硬切- 不用淡入淡出、不用柔和转场\n▍文字包装（贯穿全片）- 文字是画面主体的一部分，有真实图层关系- 文字可出现在人物前景/中景/背景之间- 人物会遮挡文字，文字也可压过人物局部，但不能遮挡眼睛和主要面部表情- 文字跟随 vocal 重音、snare、hi-hat roll、808 bass hit 出现/震动/拉伸/压缩/错位/撕裂/硬切重组- 风格：粗重压缩无衬线、巨大英文块字、竖向拉伸、横向压缩、纵排英文、弧形标题、黑白红高反差、复印纸颗粒、半色调网点、印刷破损、扫描错位、zine 拼贴感- 文字数量克制：每镜头只 1 个主文字元素，最多少量编号；不要密集小字、不要 logo\n▍律动规则（贯穿全片）- hi-hat roll → 快速微震、跳帧、文字细碎重组- snare → 文字猛然放大、画面硬切、人物肩膀下压- 808 bass hit → 低频压屏、画面短暂变形、文字纵向拉伸或横向压缩- vocal 关键词 → 嘴型同步、下颌动作、头部点拍、手势推进- 人物动作、文字动效、场景切换、景别切换必须共同卡点\n▍场景切换（贯穿全片）- 多个近景场景快速硬切- 场景气质统一参考 @图片3，但每个镜头空间要明显不同：狭窄通道、墙面近景、门口阴影、金属/混凝土背景、灯光边缘、海报墙、楼梯转角、低矮天花、车库感空间、地下入口- 不做大远景，不做复杂大场面- 每次切换由 bass hit、snare、vocal 重音或文字压屏触发\n▍景别切换（贯穿全片）- 变化要非常明显：面部极近特写 → 中近景半身 → 手部特写 → 肩颈特写 → 双人近景分屏 → 面部特写 → 半身局部- 不连续使用同一种景别- 每个硬切都让镜头距离/人物朝向/背景空间/文字层级产生明显变化- 镜头可轻微手持、突然推近、短促横移、快速压缩变焦、近距离摇摆- 不变成柔和慢镜头\n▍人物规则（贯穿全片）- 角色：两位 fly detectives / brothers- 保持 @图片1 人物形象的脸、发型、服装轮廓、比例、气质、整体氛围- 真实哑光皮肤、干净高级、自然毛孔、轻微皮肤纹理- 不要油亮 AI 美颜脸- 兄弟搭档式轮流 rap，对镜头有强烈 performance\n▍人物表演（贯穿全片）- 声音冷酷、攻击性、松弛但有压迫感，节奏紧、断裂、强拍- 不是静态摆拍，是在表演- 清楚嘴型同步：唇形、下颌、表情、呼吸跟随 vocal- 卡点动作：头部点拍、肩膀下压、手势推进、身体前倾、侧头、转身靠近镜头、手指指向镜头\n▍重要限制（贯穿全片）- 不用全身镜头- 不用多人全景- 不用复杂群像大场面- 全片只用近景/中近景/面部特写/手部特写/肩颈特写/半身镜头- 只表现人物上半身、脸部、手势、衣服材质、表情、嘴型同步、文字包装- 不要求看到完整身体和脚- 不出现真实品牌、原参考图中的 logo、标题或可识别原文字",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787573698485,
-    "promptEn": "underground rap duo freestyling at the camera",
-    "lastPlayedAt": 1788330586458,
-    "playCount": 22
-  },
-  {
-    "id": "e5246eae814e",
-    "source": "h3console",
-    "prompt": "男子道歉说自己被外面的东西追",
-    "fullPrompt": "中式古风恐怖场景，昏暗的旧屋内，油灯微弱的光照亮一间陈旧的房间，屏风后隐约可见白衣女子的身影。一位身穿古装的男子对不起，我不是有意闯进来的，外面有东西在追我。。画面氛围诡异，光影交错，充满悬疑感。电影质感，4K画质。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787569314178,
-    "promptEn": "man apologizes, says something outside was chasing him",
-    "lastPlayedAt": 1788331479210,
-    "playCount": 22
-  },
-  {
-    "id": "a0d7a106b8d6",
-    "source": "h3console",
-    "prompt": "男子道歉说自己被外面的东西追",
-    "fullPrompt": "中式古风恐怖场景，昏暗的旧屋内，油灯微弱的光照亮一间陈旧的房间，屏风后隐约可见白衣女子的身影。一位身穿古装的男子对不起，我不是有意闯进来的，外面有东西在追我。。画面氛围诡异，光影交错，充满悬疑感。电影质感，4K画质。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787568512602,
-    "lastPlayedAt": 1788330268217,
-    "promptEn": "man apologizes, says something outside was chasing him",
-    "playCount": 22
-  },
-  {
-    "id": "d71c09a38248",
-    "source": "h3console",
-    "prompt": "两个角色坐着，游戏菜单UI换装机械手臂",
-    "fullPrompt": "- 角色1参考图1,角色 2 参考图 2，UI风格参考图3。 [0 秒-2 秒] 高角度俯拍。角色1，角色 2坐在高饱和度的亮紫色地面上，参考为图1 和图 2，抬头看向摄像机。右侧显示游戏菜单 UI：开始新游戏 (START NEW GAME)、继续游戏 (CONTINUE, 高亮显示)、设置 (SETTINGS)、退出游戏 (EXIT GAME)。左上角显示玩家资料 MINIMAX。光标点击“继续游戏”。 [2 秒-4 秒] 平滑变焦至她的右臂。UI 面板从右侧滑入，“右臂装备 (RIGHT ARM EQUIPMENT)” 面板出现。选中高亮显示 “幻影之握 (PHANTOM GRIP)”，随后滑动至 “时空之爪 (CHRONOS CLAW)”。她的右手机械地重新配置，手指分开，新的爪状指节锁定到位，青色 LED 灯闪烁得更亮。 [4 秒-7 秒] 摄像机平滑地绕到她的左侧。新的 UI 滑入 “武器定制 (ARMAMENT CUSTOMIZATION)” 网格，显示手部、前臂、肘部、上臂组件。选区快速循环切换部件。她的左臂逐段拆解：前臂板脱落，新装甲滑入，肘关节更换，手部重新配置，更换过程中可见暴露的线路和活塞。 [7 秒-8.5 秒] 摄像机拉回至中景。确认配置 (CONFIRM CONFIG) 按钮闪烁。点击。所有 UI 面板向内收缩消失。角色双腿解开，调整姿势，现在放松地坐着，单膝立起，义肢手自然抬起并完成配置后的轻微动作展示。 [8.5 秒-10 秒] 底部出现加载 (LOADING) 条。从 0% 快速填充至 100%。高饱和度的亮紫色环境开始变暗，阴影从边缘蔓延，温暖的金色光芒渗入。 [10 秒-15 秒] 当她起身时，完整的环境在她周围加载。密集的赛博朋克贫民窟显现，霓虹灯闪烁，潮湿的街道反射着光芒，人群涌动，摩托车穿梭，错综复杂的电线，堆叠的建筑向远处的未来派摩天大楼延伸。摄像机稳定在第三人称视角，位于她身后，展示她的背部。HUD 元素淡入，右上角显示小地图，左下角显示生命值和弹药计数器。任务标记出现。她迈步走进街道。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787559156410,
-    "lastPlayedAt": 1788326183366,
-    "promptEn": "two characters sitting, game menu UI swaps in a mechanical arm",
-    "playCount": 21
-  },
-  {
-    "id": "87c065ab0f98",
-    "source": "h3console",
-    "prompt": "按参考视频的剪辑节奏做的片段",
-    "fullPrompt": "- 参考剪辑\n- 图1图2图3图4图5图6，严格参考示例视频视频1的镜头节奏、转场风格及音乐。",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787542770522,
-    "promptEn": "clip cut to match a reference video's editing rhythm",
-    "lastPlayedAt": 1788331138922,
-    "playCount": 22
-  },
-  {
-    "id": "f9286d694bad",
-    "source": "h3console",
-    "prompt": "乙女游戏男主角展示PV",
-    "fullPrompt": "- 生成一支 乙游男主角色 PV。 角色参考图： 使用图2作为严格的角色身份参考。请保持同一个角色脸、发型、体型比例、服装设计、材质细节和乙游 CG 质感一致。 ",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787497396594,
-    "promptEn": "otome game male lead showcase PV",
-    "lastPlayedAt": 1788327652578,
-    "playCount": 21
-  },
-  {
-    "id": "36ac12f2f29e",
-    "source": "h3console",
-    "prompt": "游戏菜单UI，角色换装机械手臂",
-    "fullPrompt": "- 角色参考图1，UI风格参考图2。 [0 秒-2 秒] 高角度俯拍。角色坐在高饱和度的亮紫色地面上，参考为图1，抬头看向摄像机。右侧显示游戏菜单 UI：开始新游戏 (START NEW GAME)、继续游戏 (CONTINUE, 高亮显示)、设置 (SETTINGS)、退出游戏 (EXIT GAME)。左上角显示玩家资料 MINIMAX。光标点击“继续游戏”。 [2 秒-4 秒] 平滑变焦至她的右臂。UI 面板从右侧滑入，“右臂装备 (RIGHT ARM EQUIPMENT)” 面板出现。选中高亮显示 “幻影之握 (PHANTOM GRIP)”，随后滑动至 “时空之爪 (CHRONOS CLAW)”。她的右手机械地重新配置，手指分开，新的爪状指节锁定到位，青色 LED 灯闪烁得更亮。 [4 秒-7 秒] 摄像机平滑地绕到她的左侧。新的 UI 滑入 “武器定制 (ARMAMENT CUSTOMIZATION)” 网格，显示手部、前臂、肘部、上臂组件。选区快速循环切换部件。她的左臂逐段拆解：前臂板脱落，新装甲滑入，肘关节更换，手部重新配置，更换过程中可见暴露的线路和活塞。 [7 秒-8.5 秒] 摄像机拉回至中景。确认配置 (CONFIRM CONFIG) 按钮闪烁。点击。所有 UI 面板向内收缩消失。角色双腿解开，调整姿势，现在放松地坐着，单膝立起，义肢手自然抬起并完成配置后的轻微动作展示。 [8.5 秒-10 秒] 底部出现加载 (LOADING) 条。从 0% 快速填充至 100%。高饱和度的亮紫色环境开始变暗，阴影从边缘蔓延，温暖的金色光芒渗入。 [10 秒-15 秒] 当她起身时，完整的环境在她周围加载。密集的赛博朋克贫民窟显现，霓虹灯闪烁，潮湿的街道反射着光芒，人群涌动，摩托车穿梭，错综复杂的电线，堆叠的建筑向远处的未来派摩天大楼延伸。摄像机稳定在第三人称视角，位于她身后，展示她的背部。HUD 元素淡入，右上角显示小地图，左下角显示生命值和弹药计数器。任务标记出现。她迈步走进街道。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787493339415,
-    "promptEn": "game menu UI, character swaps out a mechanical arm",
-    "lastPlayedAt": 1788327622554,
-    "playCount": 21
-  },
-  {
-    "id": "656c897b04fc",
-    "source": "h3console",
-    "prompt": "游戏菜单UI，角色换装机械手臂",
-    "fullPrompt": "- 角色参考图1，UI风格参考图2。 [0 秒-2 秒] 高角度俯拍。角色坐在高饱和度的亮紫色地面上，参考为图1，抬头看向摄像机。右侧显示游戏菜单 UI：开始新游戏 (START NEW GAME)、继续游戏 (CONTINUE, 高亮显示)、设置 (SETTINGS)、退出游戏 (EXIT GAME)。左上角显示玩家资料 MINIMAX。光标点击“继续游戏”。 [2 秒-4 秒] 平滑变焦至她的右臂。UI 面板从右侧滑入，“右臂装备 (RIGHT ARM EQUIPMENT)” 面板出现。选中高亮显示 “幻影之握 (PHANTOM GRIP)”，随后滑动至 “时空之爪 (CHRONOS CLAW)”。她的右手机械地重新配置，手指分开，新的爪状指节锁定到位，青色 LED 灯闪烁得更亮。 [4 秒-7 秒] 摄像机平滑地绕到她的左侧。新的 UI 滑入 “武器定制 (ARMAMENT CUSTOMIZATION)” 网格，显示手部、前臂、肘部、上臂组件。选区快速循环切换部件。她的左臂逐段拆解：前臂板脱落，新装甲滑入，肘关节更换，手部重新配置，更换过程中可见暴露的线路和活塞。 [7 秒-8.5 秒] 摄像机拉回至中景。确认配置 (CONFIRM CONFIG) 按钮闪烁。点击。所有 UI 面板向内收缩消失。角色双腿解开，调整姿势，现在放松地坐着，单膝立起，义肢手自然抬起并完成配置后的轻微动作展示。 [8.5 秒-10 秒] 底部出现加载 (LOADING) 条。从 0% 快速填充至 100%。高饱和度的亮紫色环境开始变暗，阴影从边缘蔓延，温暖的金色光芒渗入。 [10 秒-15 秒] 当她起身时，完整的环境在她周围加载。密集的赛博朋克贫民窟显现，霓虹灯闪烁，潮湿的街道反射着光芒，人群涌动，摩托车穿梭，错综复杂的电线，堆叠的建筑向远处的未来派摩天大楼延伸。摄像机稳定在第三人称视角，位于她身后，展示她的背部。HUD 元素淡入，右上角显示小地图，左下角显示生命值和弹药计数器。任务标记出现。她迈步走进街道。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787493303602,
-    "promptEn": "game menu UI, character swaps out a mechanical arm",
-    "lastPlayedAt": 1788330216170,
-    "playCount": 22
-  },
-  {
-    "id": "e1fd99e4ab97",
-    "source": "h3console",
-    "prompt": "可爱手绘小比格犬做鬼脸的游戏画面",
-    "fullPrompt": "生成一段10秒、4:3 横向构图的游戏视频，整体像一个极简可爱的 2D 手绘反应游戏界面。画面保持固定镜头，不要运镜，不要缩放，背景干净，保留温暖纸张质感和蜡笔/彩铅手绘线条。主角始终是同一只皮皮的小比格犬：圆润 Q 版身体、短腿、大耷拉耳、圆润长吻部、短翘尾巴、半眯慵懒又欠揍的表情。小比格身体可以像橡皮筋一样被夸张拉伸、压扁、拱起和侧闪，但身份和画风始终一致。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787470101960,
-    "promptEn": "cute hand-drawn beagle dog making faces, game screen",
-    "lastPlayedAt": 1788329231345,
-    "playCount": 21
-  },
-  {
-    "id": "4067993ff824",
-    "source": "h3console",
-    "prompt": "黑色背景产品发布风格展示一只烤鸭",
-    "fullPrompt": "15 秒，16:9 横屏，完整模仿高级感品牌新品发布视频的视觉语言：纯黑背景、棚拍级灯光、超慢速旋转特写、极简巨型无衬线标题、空灵电子配乐。产品是——一只完美烤制的北京烤鸭。字体必须极简、留白充足、统一使用纤细无衬线字体；中文不能有错别字。禁止：卡通风格、杂乱背景、水印。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787470101693,
-    "promptEn": "black-background product-launch style shot of a roast duck",
-    "lastPlayedAt": 1788330220175,
-    "playCount": 22
-  },
-  {
-    "id": "99e771bbb50d",
-    "source": "h3console",
-    "prompt": "竖屏家庭吵架短剧，跟中年女人对峙",
-    "fullPrompt": "室内家庭争吵短剧场景，竖屏短剧质感，真实真人表演，中式家庭/小饭馆室内环境，暖色灯光，背景有红色装饰和书法字幅，浅景深，情绪强烈，剪辑节奏紧凑。表演要求：真实短剧表演风格，不要夸张舞台腔。秦皓瑄的语气是愤怒、委屈、急切的反驳；中老年女性的语气是尖锐、强势、咄咄逼人的质问。两人之间有强烈对峙感，节奏逐步升级。画面风格：竖屏9:16，手机短剧质感，真人实拍感，浅景深，室内暖光，中近景为主，频繁正反打剪辑，背景保持生活化，不要科幻、不要古装、不要动画感。画面中不要出现任何平台水印或贴片。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787470101411,
-    "promptEn": "vertical family-fight drama, arguing with an older woman",
-    "lastPlayedAt": 1788330410331,
-    "playCount": 22
-  },
-  {
-    "id": "57169e6751d0",
-    "source": "h3console",
-    "prompt": "雪夜竹林，古装武侠悬疑对峙",
-    "fullPrompt": "古装武侠悬疑短剧场景，16:9 横向电影画幅，夜晚竹林，冷蓝、墨绿、灰黑低饱和色调，薄雾弥漫，细雪在空气中飘落。整体氛围清冷、肃杀、克制，有江湖门派办案、秘密情报交接的紧张感。画面真实古装剧质感，浅景深，电影级光影，不要现代元素，不要字幕，不要文字，不要水印。场景：夜晚竹林深处，背景是密集竖向竹竿，远处有冷白雾光透出。前景有虚化竹叶遮挡，形成“林中窥视”的空间层次。冷白侧前方柔光照亮人物面部，背景逆光营造前暗后亮的电影感，浅景深，竹叶、雪花和背景竹林虚化成柔和散景。画面风格：古装武侠电影感，冷蓝墨绿雪夜竹林，浅景深，前景竹叶遮挡，人物面部特写为主，正反打剪辑，节奏克制但紧张。不要字幕、不要平台水印、不要现代服装、不要现代建筑、不要动画感、不要过度磨皮、不要明亮白天、不要喜剧表演。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787470101060,
-    "promptEn": "snowy bamboo forest at night, wuxia mystery standoff",
-    "lastPlayedAt": 1788326978046,
-    "playCount": 21
-  },
-  {
-    "id": "125751cac1c8",
-    "source": "h3console",
-    "prompt": "全黑背景中文动态字体实验短片",
-    "fullPrompt": "整体为完全黑色背景上的中文动态字体实验影像。画面中只允许出现白色中文文字，以及极少量深红色文字错位或强调。不得出现人物、风景、照片、插画、图标、商品、Logo、3D 物体、英文标题、固定 UI、编号、计时器、边框、网格或任何装饰性图形。这不是普通字幕视频，而是以“中文文字本身”为唯一主角的高密度动态排版短片。文字在漆黑平面上高速出现、放大、旋转、压缩、碰撞、分解、增殖、反转、重组。整体风格像平面设计师制作的中文动态字体实验影像、文字运动 showreel、音乐活动开场视觉。气质黑暗、冷静、锋利、理性、无机质，略带攻击性，但保持高级、克制、统一。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787470100377,
-    "promptEn": "all-black background, experimental Chinese kinetic typography",
-    "lastPlayedAt": 1788330792633,
-    "playCount": 22
-  },
-  {
-    "id": "c24c15736daa",
-    "source": "h3console",
-    "prompt": "MiniMax H3品牌动态图形短片，圆点连接成网络",
-    "fullPrompt": "为 AI 品牌“MiniMax H3”创作一支15秒高端平面动态图形品牌影片。【必须准确显示的文字】品牌名称：MiniMax H3；品牌 Slogan：Intelligence With Everyone。除以上两组文字外，画面中不得出现任何其他可读文字。【核心创意】“One becomes everyone.” 智能从一个微小节点出发，通过连接、共享与共同生长，形成属于每个人的开放智能网络。以三种基础元素建立一套原创品牌视觉系统：一个极小的二维实心圆点／字母“H”所代表的连接桥梁／数字“3”所代表的三层智能轨道。所有圆点、线条、轨道、字形和转场必须由这套基础元素连续演化。每次变化都具有清晰的视觉因果关系，形成统一的品牌视觉 DNA。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787470100107,
-    "promptEn": "MiniMax H3 brand motion graphics, a dot growing into a network",
-    "lastPlayedAt": 1788328939194,
-    "playCount": 21
-  },
-  {
-    "id": "194e9bd77da6",
-    "source": "h3console",
-    "prompt": "MiniMax H3品牌动态图形短片，圆点连接成网络",
-    "fullPrompt": "为 AI 品牌“MiniMax H3”创作一支15秒高端平面动态图形品牌影片。【必须准确显示的文字】品牌名称：MiniMax H3；品牌 Slogan：Intelligence With Everyone。除以上两组文字外，画面中不得出现任何其他可读文字。【核心创意】“One becomes everyone.” 智能从一个微小节点出发，通过连接、共享与共同生长，形成属于每个人的开放智能网络。以三种基础元素建立一套原创品牌视觉系统：一个极小的二维实心圆点／字母“H”所代表的连接桥梁／数字“3”所代表的三层智能轨道。所有圆点、线条、轨道、字形和转场必须由这套基础元素连续演化。每次变化都具有清晰的视觉因果关系，形成统一的品牌视觉 DNA。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787470100022,
-    "promptEn": "MiniMax H3 brand motion graphics, a dot growing into a network",
-    "lastPlayedAt": 1788330698554,
-    "playCount": 22
-  },
-  {
-    "id": "eaf2a1433c41",
-    "source": "h3console",
-    "prompt": "手指向海边集市的物品，变成手绘动画",
-    "fullPrompt": "视频整体 16:9 横向视频，总时长约 10～15 秒，全 4 个镜头，每个镜头约 3 秒，用节奏明快的硬切连接。画面是“海边小集市”的实拍智能手机影像，背景包含浅色摊位、遮阳棚、贝壳饰品、冰饮摊、海风吹动的布帘、远处沙滩和淡蓝色海面，整体氛围明亮、轻快、夏日感强。拍摄者右手以正常、自然、正确的写字握笔姿势拿一支细的青蓝色笔，从画面下部进入，每个镜头只指向一个现实对象，被指到的对象整体变成粗糙手绘 2D 动画，其他背景和未指向元素全部保持实拍。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787470099635,
-    "promptEn": "pointing at things at a beach market, they turn into hand-drawn animation",
-    "lastPlayedAt": 1788329045254,
-    "playCount": 21
-  },
-  {
-    "id": "0fe0f2101c44",
-    "source": "h3console",
-    "prompt": "深夜洗衣店实拍混手绘发光动画",
-    "fullPrompt": "15 秒、16:9 横版短视频。实拍深夜自助洗衣店场景，融合手绘发光动画的混合影像。一间小型自助洗衣店，荧光灯微微频闪，店内摆放着运转中的洗衣机、塑料洗衣篮、老旧长椅，地面散落一只袜子，整体空间安静，带着淡淡的怀旧氛围感。手持手机单手拍摄质感，画面抖动明显；白色荧光灯造成曝光忽明忽暗；玻璃表面带有环境反光；镜头靠近物体时对焦存在延迟。画面不要像商业广告一样精致规整，整体质感如同深夜偶然闯入、追着奇异幻象随手拍下的真实纪实感。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787470099306,
-    "promptEn": "late-night laundromat footage mixed with glowing hand-drawn animation",
-    "lastPlayedAt": 1788329245355,
-    "playCount": 21
-  },
-  {
-    "id": "c0e23beb4132",
-    "source": "h3console",
-    "prompt": "傍晚小厨房实拍混手绘发光生物动画",
-    "fullPrompt": "15秒、16:9横向视频。将傍晚小厨房的真人实拍与手绘发光动画融合在一起的影像。夕阳余晖还残留在窗边，生活感十足的小厨房里有一张旧木桌、洗到一半的马克杯、略微起雾的玻璃瓶，以及悬挂着的抹布。画面带有像用智能手机单手拍摄一样的细微手抖、近距离对焦时的犹豫、逆光造成的曝光波动，以及暗部略显粗糙的噪点。不要像广告影像那样精心整理，而是要呈现出像是在家中慌忙拍下某个不可思议事件的自然质感。禁止出现巨大的眼睛、裂开的嘴、獠牙、威吓、扑袭动作、突然黑屏和跳吓。声音只使用厨房室内环境声、抹布摩擦声、马克杯轻微碰撞声、水龙头水滴声、拍摄者的脚步声与轻微呼吸声，以及手绘生物柔和的电子音和小小的叫声。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787470098914,
-    "promptEn": "evening kitchen footage mixed with a glowing hand-drawn creature",
-    "lastPlayedAt": 1788330722576,
-    "playCount": 22
-  },
-  {
-    "id": "6639b07acc7b",
-    "source": "h3console",
-    "prompt": "快速硬切的宇宙飞船预告片头字幕",
-    "fullPrompt": "节奏更快，宏大但不拖。快速硬切、舰桥震动、强光爆闪、短黑场、跃迁冲击。文字是电影预告片式宽字距标题包装，字体不要纯白，要有克制发光和材质感、微弱边缘辉光。文字动效包含从深空暗部渐入、被星光扫亮、字距展开、残影、微弱发光、黑场闪现。",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787470098570,
-    "promptEn": "fast-cut spaceship trailer title sequence",
-    "lastPlayedAt": 1788331317067,
-    "playCount": 22
-  },
-  {
-    "id": "1e55e540df11",
-    "source": "h3console",
-    "prompt": "黎明宁静山村（第二次尝试）",
-    "fullPrompt": "a quiet mountain village at dawn, second attempt",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787461994394,
-    "promptEn": "a quiet mountain village at dawn (second attempt)",
-    "lastPlayedAt": 1788327028088,
-    "playCount": 21
-  },
-  {
-    "id": "8e16b6faafd9",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 448,
-    "height": 256,
-    "duration": 125,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787455895150,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788326735846,
-    "playCount": 21
-  },
-  {
-    "id": "2a4e45820135",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 448,
-    "height": 256,
-    "duration": 112.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787453697093,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788329361450,
-    "playCount": 21
-  },
-  {
-    "id": "bfe13d1cc4a2",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 448,
-    "height": 256,
-    "duration": 75,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787452521817,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788331359103,
-    "playCount": 22
-  },
-  {
-    "id": "f36cdd73ff1b",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 448,
-    "height": 256,
-    "duration": 50,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787450754171,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788328797084,
-    "playCount": 21
-  },
-  {
-    "id": "fd0183c54763",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 17.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787447175021,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788327388389,
-    "playCount": 21
-  },
-  {
-    "id": "4580ca66725a",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 15.6,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787444313690,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788329765778,
-    "playCount": 22
-  },
-  {
-    "id": "e56962bd770f",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787442174982,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788330522408,
-    "playCount": 22
-  },
-  {
-    "id": "18105bc2900b",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 46.9,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787437988124,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788327928809,
-    "playCount": 21
-  },
-  {
-    "id": "e497093c79ce",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 31.3,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787436029475,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788329779791,
-    "playCount": 22
-  },
-  {
-    "id": "896859e59dae",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 31,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787432295531,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788330184146,
-    "playCount": 22
-  },
-  {
-    "id": "79368c9cf499",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 29.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787428860940,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788329951946,
-    "playCount": 22
-  },
-  {
-    "id": "dbc7c27b72ba",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 20.8,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787426955312,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788327596538,
-    "playCount": 21
-  },
-  {
-    "id": "734d28c52670",
-    "source": "h3console",
-    "prompt": "小船在晨雾森林河道中行驶",
-    "fullPrompt": "a small boat sailing on a calm river through a forest, morning mist",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 8.3,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787426484939,
-    "promptEn": "small boat sailing through a misty forest river",
-    "lastPlayedAt": 1788327428425,
-    "playCount": 21
-  },
-  {
-    "id": "9fbcf0d87021",
-    "source": "h3console",
-    "prompt": "悬崖灯塔前夕阳落日，海鸥飞过",
-    "fullPrompt": "a lighthouse on a rocky cliff at sunset, waves crashing, seagulls flying",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 8.3,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787425932119,
-    "promptEn": "lighthouse on a cliff at sunset, seagulls flying",
-    "lastPlayedAt": 1788328490844,
-    "playCount": 21
-  },
-  {
-    "id": "cf2015093ec4",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 20.8,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787415067615,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788329981970,
-    "playCount": 22
-  },
-  {
-    "id": "9bcff387cddf",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 16.7,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787413999418,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788326950031,
-    "playCount": 21
-  },
-  {
-    "id": "ade0318a7a4a",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 16.7,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787413511959,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788327110160,
-    "playCount": 21
-  },
-  {
-    "id": "5d4f40bf0889",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 12.5,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787412952086,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788330078049,
-    "playCount": 22
-  },
-  {
-    "id": "59fd7a087f5b",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 8.3,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787412606465,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788327020082,
-    "playCount": 21
-  },
-  {
-    "id": "29a99dc412b2",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 2.3,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787412150062,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788331028834,
-    "playCount": 22
-  },
-  {
-    "id": "9d2486d6f969",
-    "source": "h3console",
-    "prompt": "一只猫坐在窗台上",
-    "fullPrompt": "A cat sitting on a windowsill.",
-    "edited": true,
-    "width": 320,
-    "height": 320,
-    "duration": 0.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787397450973,
-    "promptEn": "a cat sitting on a windowsill",
-    "lastPlayedAt": 1788329847850,
-    "playCount": 22
-  },
-  {
-    "id": "580d1d722adc",
-    "source": "h3console",
-    "prompt": "一只猫坐在窗台上",
-    "fullPrompt": "A cat sitting on a windowsill.",
-    "edited": true,
-    "width": 320,
-    "height": 320,
-    "duration": 0.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787388090646,
-    "promptEn": "a cat sitting on a windowsill",
-    "lastPlayedAt": 1788330846675,
-    "playCount": 22
-  },
-  {
-    "id": "c551457bc55e",
-    "source": "h3console",
-    "prompt": "一只猫坐在窗台上",
-    "fullPrompt": "A cat sitting on a windowsill.",
-    "edited": true,
-    "width": 320,
-    "height": 320,
-    "duration": 0.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787386316175,
-    "lastPlayedAt": 1788330660519,
-    "promptEn": "a cat sitting on a windowsill",
-    "playCount": 22
-  },
-  {
-    "id": "f9b68e5daaed",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 2.3,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787384983124,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788328684991,
-    "playCount": 21
-  },
-  {
-    "id": "7753a046ba79",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787384417941,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788331164936,
-    "playCount": 22
-  },
-  {
-    "id": "2cf5c8e67510",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 1024,
-    "height": 576,
-    "duration": 1.3,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787383911831,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788326235412,
-    "playCount": 21
-  },
-  {
-    "id": "9cba879e263d",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one buildings roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 448,
-    "height": 256,
-    "duration": 10.1,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787375413935,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788331108901,
-    "playCount": 22
-  },
-  {
-    "id": "10112fb3c1fd",
-    "source": "h3console",
-    "prompt": "金色传送门打开，绿夹克男人扔出发光长矛",
-    "fullPrompt": "A golden portal opens in the night sky above a rooftop plaza, a man in a green jacket throws a glowing spear toward the camera, cinematic lighting, smooth continuous motion.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 2.7,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787375413623,
-    "promptEn": "golden portal opens, man in green jacket throws a glowing spear",
-    "lastPlayedAt": 1788330666526,
-    "playCount": 22
-  },
-  {
-    "id": "78e698c924b1",
-    "source": "h3console",
-    "prompt": "金色传送门打开，绿夹克男人扔出发光长矛",
-    "fullPrompt": "A golden portal opens in the night sky above a rooftop plaza, a man in a green jacket throws a glowing spear toward the camera, cinematic lighting, smooth continuous motion.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 2.7,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787375413203,
-    "promptEn": "golden portal opens, man in green jacket throws a glowing spear",
-    "lastPlayedAt": 1788329243353,
-    "playCount": 21
-  },
-  {
-    "id": "219ed88c3ea7",
-    "source": "h3console",
-    "prompt": "金色传送门打开，绿夹克男人扔出发光长矛",
-    "fullPrompt": "A golden portal opens in the night sky above a rooftop plaza, a man in a green jacket throws a glowing spear toward the camera, cinematic lighting, smooth continuous motion.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 2.7,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787375412946,
-    "promptEn": "golden portal opens, man in green jacket throws a glowing spear",
-    "lastPlayedAt": 1788327562526,
-    "playCount": 21
-  },
-  {
-    "id": "a72236902363",
-    "source": "h3console",
-    "prompt": "金色传送门打开，绿夹克男人扔出发光长矛",
-    "fullPrompt": "A golden portal opens in the night sky above a rooftop plaza, a man in a green jacket throws a glowing spear toward the camera, cinematic lighting, smooth continuous motion.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 2.7,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787367454249,
-    "promptEn": "golden portal opens, man in green jacket throws a glowing spear",
-    "lastPlayedAt": 1788331485209,
-    "playCount": 22
-  },
-  {
-    "id": "c48af2f34e62",
-    "source": "h3console",
-    "prompt": "金色传送门打开，绿夹克男人扔出发光长矛",
-    "fullPrompt": "A golden portal opens in the night sky above a rooftop plaza, a man in a green jacket throws a glowing spear toward the camera, cinematic lighting, smooth continuous motion.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 2.7,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787367454065,
-    "promptEn": "golden portal opens, man in green jacket throws a glowing spear",
-    "lastPlayedAt": 1788328997227,
-    "playCount": 21
-  },
-  {
-    "id": "e6c241be377f",
-    "source": "h3console",
-    "prompt": "金色传送门打开，绿夹克男人扔出发光长矛",
-    "fullPrompt": "A golden portal opens in the night sky above a rooftop plaza, a man in a green jacket throws a glowing spear toward the camera, cinematic lighting, smooth continuous motion.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 2.7,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787367453839,
-    "promptEn": "golden portal opens, man in green jacket throws a glowing spear",
-    "lastPlayedAt": 1788330844671,
-    "playCount": 22
-  },
-  {
-    "id": "8402adae59dc",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one building's roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 448,
-    "height": 256,
-    "duration": 10.1,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787363956909,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788329891890,
-    "playCount": 22
-  },
-  {
-    "id": "5692965402f5",
-    "source": "h3console",
-    "prompt": "金色传送门打开，绿夹克男人扔出发光长矛",
-    "fullPrompt": "A golden portal opens in the night sky above a rooftop plaza, a man in a green jacket throws a glowing spear toward the camera, cinematic lighting, smooth continuous motion.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 2.7,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787332996051,
-    "promptEn": "golden portal opens, man in green jacket throws a glowing spear",
-    "lastPlayedAt": 1788327832724,
-    "playCount": 21
-  },
-  {
-    "id": "bf7d973d02a3",
-    "source": "h3console",
-    "prompt": "金色传送门打开，绿夹克男人扔出发光长矛",
-    "fullPrompt": "A golden portal opens in the night sky above a rooftop plaza, a man in a green jacket throws a glowing spear toward the camera, cinematic lighting, smooth continuous motion.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 2.7,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787332047061,
-    "promptEn": "golden portal opens, man in green jacket throws a glowing spear",
-    "lastPlayedAt": 1788329537599,
-    "playCount": 21
-  },
-  {
-    "id": "a5688d8f5db3",
-    "source": "h3console",
-    "prompt": "木桌上一个红苹果，电影打光",
-    "fullPrompt": "a red apple on a wooden table, cinematic lighting",
-    "edited": true,
-    "width": 320,
-    "height": 320,
-    "duration": 0.2,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787330101793,
-    "promptEn": "a red apple on a wooden table, cinematic lighting",
-    "lastPlayedAt": 1788330428348,
-    "playCount": 22
-  },
-  {
-    "id": "0024d3c2d4cb",
-    "source": "h3console",
-    "prompt": "江南古镇晨雾中，女子撑伞走过青石巷",
-    "fullPrompt": "清晨薄雾中的江南古镇，青石板小巷，一位撑着油纸伞的女子缓缓走过，远处有船夫摇橹划过河面，光影随风吹拂的柳枝轻轻晃动",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787326019663,
-    "promptEn": "misty Jiangnan old town at dawn, woman walking with an oil-paper umbrella",
-    "lastPlayedAt": 1788330558436,
-    "playCount": 22
-  },
-  {
-    "id": "ab42a6fba515",
-    "source": "h3console",
-    "prompt": "江南古镇晨雾中，女子撑伞走过青石巷",
-    "fullPrompt": "清晨薄雾中的江南古镇，青石板小巷，一位撑着油纸伞的女子缓缓走过，远处有船夫摇橹划过河面，光影随风吹拂的柳枝轻轻晃动",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787326019290,
-    "promptEn": "misty Jiangnan old town at dawn, woman walking with an oil-paper umbrella",
-    "lastPlayedAt": 1788327762673,
-    "playCount": 21
-  },
-  {
-    "id": "fd8534ce588c",
-    "source": "h3console",
-    "prompt": "透明游戏鼠标产品片，蓝橙灯光",
-    "fullPrompt": "Editorial tech product film. The transparent gaming mouse from <Picture 1> in its original scene: a pitch-black studio void with a dark, subtle reflective surface, lit by dramatic duotone vibrant blue and warm neon orange rim lighting, deep soft shadow falloff into pure black. Monochromatic dark palette with electric blue and amber accents. Material motif: glowing internal metallic micro-components and glossy acrylic refractions. The environment is constant throughout.\nSHOT 1: The scene opens exactly on image 1, the mouse resting confidently on the dark surface; the blue and orange lights slowly pulse brighter, refracting deeply through the transparent acrylic shell as the camera executes a slow, deliberate push-in to reveal the intricate circuitry.\nSHOT 2: Cut to an extreme macro profile of the ridged scroll wheel and layered internal micro-components; the camera glides slowly along the side as a sharp beam of warm orange light sweeps across the metallic textures, contrasting perfectly against the deep blue ambient glow.\nSHOT 3: Cut to a low-angle beauty shot: the mouse levitates weightlessly a few centimeters above the dark reflective surface, rotating in a slow, precise orbit; the duotone lighting flares gently along the glassy transparent edges before fading slowly into a sleek silhouette.\nAudio: deep pulsing sub-bass room tone, sharp tactile mechanical clicks, a sweeping glassy whoosh on cuts, and a rising electronic swell that resolves to near-silence on the final fade.
",
-    "edited": true,
-    "width": 448,
-    "height": 256,
-    "duration": 15.1,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787309864037,
-    "promptEn": "transparent gaming mouse product film, blue and orange lighting",
-    "lastPlayedAt": 1788330988797,
-    "playCount": 22
-  },
-  {
-    "id": "073f5df6427a",
-    "source": "h3console",
-    "prompt": "透明游戏鼠标产品片，蓝橙灯光",
-    "fullPrompt": "Editorial tech product film. The transparent gaming mouse from <Picture 1> in its original scene: a pitch-black studio void with a dark, subtle reflective surface, lit by dramatic duotone vibrant blue and warm neon orange rim lighting, deep soft shadow falloff into pure black. Monochromatic dark palette with electric blue and amber accents. Material motif: glowing internal metallic micro-components and glossy acrylic refractions. The environment is constant throughout.\nSHOT 1: The scene opens exactly on image 1, the mouse resting confidently on the dark surface; the blue and orange lights slowly pulse brighter, refracting deeply through the transparent acrylic shell as the camera executes a slow, deliberate push-in to reveal the intricate circuitry.\nSHOT 2: Cut to an extreme macro profile of the ridged scroll wheel and layered internal micro-components; the camera glides slowly along the side as a sharp beam of warm orange light sweeps across the metallic textures, contrasting perfectly against the deep blue ambient glow.\nSHOT 3: Cut to a low-angle beauty shot: the mouse levitates weightlessly a few centimeters above the dark reflective surface, rotating in a slow, precise orbit; the duotone lighting flares gently along the glassy transparent edges before fading slowly into a sleek silhouette.\nAudio: deep pulsing sub-bass room tone, sharp tactile mechanical clicks, a sweeping glassy whoosh on cuts, and a rising electronic swell that resolves to near-silence on the final fade.
",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787305544815,
-    "promptEn": "transparent gaming mouse product film, blue and orange lighting",
-    "lastPlayedAt": 1788330282230,
-    "playCount": 22
-  },
-  {
-    "id": "e23b04ac2c78",
-    "source": "h3console",
-    "prompt": "主角在高楼间被追逐跳跃逃命",
-    "fullPrompt": "Realistic live-action cinematic look, action movie trailer: practical film photography style, a post-rain dusk metropolis, anamorphic lens, shallow depth of field, film grain, city volumetric fog, flying-car traffic between the towers, restrained grading for a premium feel, powerful natural movement.\n\nScene overview: at dusk on a cluster of skyscrapers, the protagonist is being chased, sprinting and leaping across rooftops, jumping from one building's roof to the next with pursuers closing in behind. This is the escape sequence of an action movie trailer: every leap is life-or-death, thrilling and fluid.\n\nStoryboard (each shot a separate scene, rapid cuts, all landing on the musical beats):\n[0s-1.5s] Shot 1: high side angle: the protagonist sprinting at the roof edge, pursuers appearing in the rooftop doorway behind him, wind catching his coat.\n[1s-2.5s] Shot 2: the protagonist leaps across the gap between buildings, body stretching mid-air, towers and flying-car light trails behind him, a slight slow-motion feel.\n[2.5s-4s] Shot 3: he lands, rolls and rises, low-angle shot, tower shadows and fog behind him, he keeps running.\n[4s-5s] Shot 4: freeze: the instant he hits the edge of the next roof and launches into the jump, silhouette, holding.\n\nCamera: each shot its own angle, cuts clean and hard, no dissolves, a slight frame jitter on the jumps.\n\nAudio: wind, rapid footsteps, city ambience, low score underneath, an accent hit on each leap, the score bursting at 4s, closing the last 1s.\n\nNo text, subtitles, logos or watermarks of any kind, no animation or cartoon rendering, no overly-CG look, keep the live-action texture.",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787304875681,
-    "promptEn": "protagonist leaps across rooftops fleeing pursuers",
-    "lastPlayedAt": 1788327912797,
-    "playCount": 21
-  },
-  {
-    "id": "c84bc1c7bfa0",
-    "source": "h3console",
-    "prompt": "捕快隔空问女子怎么确定不是人，她说人总该有声音",
-    "fullPrompt": "中国古典志怪写实电影，9秒，16:9横屏，悬疑克制。@图片1锁定首帧站位与内室，@图片2、@图片4锁服饰体态，女子与捕快面部分别由@图片3、@图片5锁定。以参考图为基准，五官、服装、陈设不变。\n0-3秒画面：严格保持@图片1构图，捕快在画外问“你怎么知道那是个‘东西’，不是个人？”，镜头越肩对准灯边女子。\n3-6秒画面：女子面部近景，女子微顿、拢袖，目光移回画外捕快；女子口型同步“若是人，总该有个人声吧？”。\n6-9秒画面：切捕快面部近景，他沉默不答；镜头拉回中景双人，油灯仍亮。\n【禁止】门、门槛、屋外、入场、推门、闯入镜头，捕快正面说话口型，揭示女子身份或美人图秘密。\n【声音】油灯噼啪、屋外低雨；对白只走音频。【字幕】无字幕；对白只走音频。\n最终帧：中景双人，捕快立在画面左侧凝视女子，女子在屏风近旁的灯光里拢着袖口，油灯仍亮\n\n",
-    "edited": true,
-    "width": 448,
-    "height": 256,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787304031214,
-    "promptEn": "constable asks how she's sure it's not human, she says a person would make a sound",
-    "lastPlayedAt": 1788327466457,
-    "playCount": 21
-  },
-  {
-    "id": "23d2fd81eb4d",
-    "source": "h3console",
-    "prompt": "捕快隔空问女子怎么确定不是人，她说人总该有声音",
-    "fullPrompt": "中国古典志怪写实电影，9秒，16:9横屏，悬疑克制。@图片1锁定首帧站位与内室，@图片2、@图片4锁服饰体态，女子与捕快面部分别由@图片3、@图片5锁定。以参考图为基准，五官、服装、陈设不变。\n0-3秒画面：严格保持@图片1构图，捕快在画外问“你怎么知道那是个‘东西’，不是个人？”，镜头越肩对准灯边女子。\n3-6秒画面：女子面部近景，女子微顿、拢袖，目光移回画外捕快；女子口型同步“若是人，总该有个人声吧？”。\n6-9秒画面：切捕快面部近景，他沉默不答；镜头拉回中景双人，油灯仍亮。\n【禁止】门、门槛、屋外、入场、推门、闯入镜头，捕快正面说话口型，揭示女子身份或美人图秘密。\n【声音】油灯噼啪、屋外低雨；对白只走音频。【字幕】无字幕；对白只走音频。\n最终帧：中景双人，捕快立在画面左侧凝视女子，女子在屏风近旁的灯光里拢着袖口，油灯仍亮\n\n",
-    "edited": true,
-    "width": 448,
-    "height": 256,
-    "duration": 15.1,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787304031103,
-    "promptEn": "constable asks how she's sure it's not human, she says a person would make a sound",
-    "lastPlayedAt": 1788330092062,
-    "playCount": 22
-  },
-  {
-    "id": "807e042dce0d",
-    "source": "h3console",
-    "prompt": "江南古镇晨雾中，女子撑伞走过青石巷",
-    "fullPrompt": "清晨薄雾中的江南古镇，青石板小巷，一位撑着油纸伞的女子缓缓走过，远处有船夫摇橹划过河面，光影随风吹拂的柳枝轻轻晃动",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787290119529,
-    "promptEn": "misty Jiangnan old town at dawn, woman walking with an oil-paper umbrella",
-    "lastPlayedAt": 1788330760611,
-    "playCount": 22
-  },
-  {
-    "id": "06bb43196c46",
-    "source": "h3console",
-    "prompt": "江南古镇晨雾中，女子撑伞走过青石巷",
-    "fullPrompt": "清晨薄雾中的江南古镇，青石板小巷，一位撑着油纸伞的女子缓缓走过，远处有船夫摇橹划过河面，光影随风吹拂的柳枝轻轻晃动",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787290119114,
-    "promptEn": "misty Jiangnan old town at dawn, woman walking with an oil-paper umbrella",
-    "lastPlayedAt": 1788328414771,
-    "playCount": 21
-  },
-  {
-    "id": "0f22388f7847",
-    "source": "h3console",
-    "prompt": "江南古镇晨雾中，女子撑伞走过青石巷",
-    "fullPrompt": "清晨薄雾中的江南古镇，青石板小巷，一位撑着油纸伞的女子缓缓走过，远处有船夫摇橹划过河面，光影随风吹拂的柳枝轻轻晃动",
-    "edited": true,
-    "width": 1344,
-    "height": 768,
-    "duration": 2.7,
-    "partition": "text",
-    "likes": 0,
-    "createdAt": 1787290118806,
-    "promptEn": "misty Jiangnan old town at dawn, woman walking with an oil-paper umbrella",
-    "lastPlayedAt": 1788331453182,
-    "playCount": 22
-  },
-  {
-    "id": "0c67d73b06d4",
-    "source": "h3console",
-    "prompt": "捕快用油灯烧毁美人图，房门自己打开",
-    "fullPrompt": "将@视频1延长10秒。从@视频1最后一帧开始：中景同框，完整美人图的皮质边缘被灯照亮，女子在灯光里可见，捕快侧脸停在两者之间。 不要重复逐字读画或同框比脸。以参考图为基准，女子、捕快面部由@图片3、@图片5锁定，服饰与内室陈设不变。 0-3秒画面：捕快手抖着把油灯火苗逼近画角，回头盯住女子，三者同框。 3-6秒画面：火苗卷上画边，皮质画幅迅速蜷曲破毁，捕快面部近景映出火光。 6-8秒画面：油灯骤灭，封闭房门自行打开；镜头随捕快回头，不拍屋外。 8-10秒画面：拉成全景，毁坏的画、熄灭的灯、打开的门与灯边女子都在画内。 【禁止】捕快跨门离屋、女子怪物化或攻击、女子正面说话口型、复原美人图、屋外画面。 【声音】火焰收紧、门轴低响；女子画外音“你明明差一点就会留下来”。【字幕】无字幕；对白只走音频。 最终帧：全景，捕快面向打开的房门停在屋内，油灯已熄，美人图已经毁坏，女子仍在原灯位后方可见",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787231456079,
-    "promptEn": "constable burns the beauty portrait with the lamp, the door opens on its own",
-    "lastPlayedAt": 1788330278228,
-    "playCount": 22
-  },
-  {
-    "id": "66158f543a20",
-    "source": "h3console",
-    "prompt": "捕快用油灯烧毁美人图，房门自己打开",
-    "fullPrompt": "将@视频1延长10秒。从@视频1最后一帧开始：中景同框，完整美人图的皮质边缘被灯照亮，女子在灯光里可见，捕快侧脸停在两者之间。 不要重复逐字读画或同框比脸。以参考图为基准，女子、捕快面部由@图片3、@图片5锁定，服饰与内室陈设不变。 0-3秒画面：捕快手抖着把油灯火苗逼近画角，回头盯住女子，三者同框。 3-6秒画面：火苗卷上画边，皮质画幅迅速蜷曲破毁，捕快面部近景映出火光。 6-8秒画面：油灯骤灭，封闭房门自行打开；镜头随捕快回头，不拍屋外。 8-10秒画面：拉成全景，毁坏的画、熄灭的灯、打开的门与灯边女子都在画内。 【禁止】捕快跨门离屋、女子怪物化或攻击、女子正面说话口型、复原美人图、屋外画面。 【声音】火焰收紧、门轴低响；女子画外音“你明明差一点就会留下来”。【字幕】无字幕；对白只走音频。 最终帧：全景，捕快面向打开的房门停在屋内，油灯已熄，美人图已经毁坏，女子仍在原灯位后方可见",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787231440149,
-    "promptEn": "constable burns the beauty portrait with the lamp, the door opens on its own",
-    "lastPlayedAt": 1788326655768,
-    "playCount": 21
-  },
-  {
-    "id": "63c91f4b6bc7",
-    "source": "h3console",
-    "prompt": "捕快用油灯烧毁美人图，房门自己打开",
-    "fullPrompt": "将@视频1延长10秒。从@视频1最后一帧开始：中景同框，完整美人图的皮质边缘被灯照亮，女子在灯光里可见，捕快侧脸停在两者之间。 不要重复逐字读画或同框比脸。以参考图为基准，女子、捕快面部由@图片3、@图片5锁定，服饰与内室陈设不变。 0-3秒画面：捕快手抖着把油灯火苗逼近画角，回头盯住女子，三者同框。 3-6秒画面：火苗卷上画边，皮质画幅迅速蜷曲破毁，捕快面部近景映出火光。 6-8秒画面：油灯骤灭，封闭房门自行打开；镜头随捕快回头，不拍屋外。 8-10秒画面：拉成全景，毁坏的画、熄灭的灯、打开的门与灯边女子都在画内。 【禁止】捕快跨门离屋、女子怪物化或攻击、女子正面说话口型、复原美人图、屋外画面。 【声音】火焰收紧、门轴低响；女子画外音“你明明差一点就会留下来”。【字幕】无字幕；对白只走音频。 最终帧：全景，捕快面向打开的房门停在屋内，油灯已熄，美人图已经毁坏，女子仍在原灯位后方可见",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787225897650,
-    "promptEn": "constable burns the beauty portrait with the lamp, the door opens on its own",
-    "lastPlayedAt": 1788328558886,
-    "playCount": 21
-  },
-  {
-    "id": "ed3663dcda7f",
-    "source": "h3console",
-    "prompt": "捕快用油灯烧毁美人图，房门自己打开",
-    "fullPrompt": "将@视频1延长10秒。从@视频1最后一帧开始：中景同框，完整美人图的皮质边缘被灯照亮，女子在灯光里可见，捕快侧脸停在两者之间。 不要重复逐字读画或同框比脸。以参考图为基准，女子、捕快面部由@图片3、@图片5锁定，服饰与内室陈设不变。 0-3秒画面：捕快手抖着把油灯火苗逼近画角，回头盯住女子，三者同框。 3-6秒画面：火苗卷上画边，皮质画幅迅速蜷曲破毁，捕快面部近景映出火光。 6-8秒画面：油灯骤灭，封闭房门自行打开；镜头随捕快回头，不拍屋外。 8-10秒画面：拉成全景，毁坏的画、熄灭的灯、打开的门与灯边女子都在画内。 【禁止】捕快跨门离屋、女子怪物化或攻击、女子正面说话口型、复原美人图、屋外画面。 【声音】火焰收紧、门轴低响；女子画外音“你明明差一点就会留下来”。【字幕】无字幕；对白只走音频。 最终帧：全景，捕快面向打开的房门停在屋内，油灯已熄，美人图已经毁坏，女子仍在原灯位后方可见",
-    "edited": true,
-    "width": 448,
-    "height": 256,
-    "duration": 15.1,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787224701504,
-    "lastPlayedAt": 1788329877879,
-    "promptEn": "constable burns the beauty portrait with the lamp, the door opens on its own",
-    "playCount": 22
-  },
-  {
-    "id": "f78ebe4c01c8",
-    "source": "h3console",
-    "prompt": "捕快用油灯烧毁美人图，房门自己打开",
-    "fullPrompt": "将@视频1延长10秒。从@视频1最后一帧开始：中景同框，完整美人图的皮质边缘被灯照亮，女子在灯光里可见，捕快侧脸停在两者之间。 不要重复逐字读画或同框比脸。以参考图为基准，女子、捕快面部由@图片3、@图片5锁定，服饰与内室陈设不变。 0-3秒画面：捕快手抖着把油灯火苗逼近画角，回头盯住女子，三者同框。 3-6秒画面：火苗卷上画边，皮质画幅迅速蜷曲破毁，捕快面部近景映出火光。 6-8秒画面：油灯骤灭，封闭房门自行打开；镜头随捕快回头，不拍屋外。 8-10秒画面：拉成全景，毁坏的画、熄灭的灯、打开的门与灯边女子都在画内。 【禁止】捕快跨门离屋、女子怪物化或攻击、女子正面说话口型、复原美人图、屋外画面。 【声音】火焰收紧、门轴低响；女子画外音“你明明差一点就会留下来”。【字幕】无字幕；对白只走音频。 最终帧：全景，捕快面向打开的房门停在屋内，油灯已熄，美人图已经毁坏，女子仍在原灯位后方可见",
-    "edited": true,
-    "width": 448,
-    "height": 256,
-    "duration": 5.2,
-    "partition": "fl2va",
-    "likes": 0,
-    "createdAt": 1787224519456,
-    "promptEn": "constable burns the beauty portrait with the lamp, the door opens on its own",
-    "lastPlayedAt": 1788331134918,
-    "playCount": 22
-  },
-  {
-    "id": "a1cc621cc6b2",
-    "source": "h3console",
-    "prompt": "哥特古堡月夜，吸血鬼贵族抱起公主，玫瑰花瓣飘落",
-    "fullPrompt": "哥特古堡月夜场景，固定中景镜头。黑发吸血鬼贵族男性身穿黑色礼服披风，将金发女性公主抱在怀中，女性身穿白色丝绸长裙。两人面庞靠近，目光对视，氛围暧昧而危险。背景是哥特式尖拱回廊，满月高悬夜空，翻涌的云层，两侧烛台燃烧暖黄烛火，石柱上缠绕暗红玫瑰藤蔓，红色花瓣漫天飘落。Live2D动画质感：夜风吹动男性的黑发和披风衣摆，衣褶随风飘动；女性的金色长发和裙摆薄纱被风向右吹拂，裙摆轻颤；两人胸口的衣料随呼吸微微起伏；女性眨眼、睫毛轻颤；男性目光深沉凝视；女性搭在男性肩上的手指微动，轻触他的后颈；红色花瓣在风中旋转飘落，一些花瓣擦过人物衣角。整体气氛浪漫、华丽、神秘，冷色调月光与暖色烛光交织，充满哥特爱情故事的宿命感。",
-    "edited": true,
-    "width": 832,
-    "height": 480,
-    "duration": 5.2,
-    "partition": "ref2va",
-    "likes": 0,
-    "createdAt": 1787220579189,
-    "promptEn": "gothic castle at night, vampire noble carries a princess, rose petals falling",
-    "lastPlayedAt": 1788331301054,
-    "playCount": 22
-  }
-]
\ No newline at end of file