TS-16.md
9.66 KB
TS-16 交付物
第 0 步判定
- 模式:参考视频(舞者形象与动作)+ 两段人声干声(音色参考)→ 全能参考(① 型六段)。
- 交付形态:① 型(六段,正文英文);全能参考无指令行。
- 时长:10 秒,落在 4–15s 内(H2 通过)→ 3 镜切分(0.000–3.500 / 3.500–6.800 / 6.800–10.000)。
- 素材总量:视频 1 + 音频 2 = 3 文件 ≤12,视频 ≤3,音频 ≤3;音频未单独输入(与参考视频同批,H13 音频不可单独输入的条件满足)。
- 台词/歌词:输入未给任何歌词 → 不臆造歌词文本,只描述演唱行为本身。
最终提示词原文
subject_definitions:
<Subject 1> is the male lead vocalist and dancer shown in <Video 1>; his face, hairstyle, build, and the exact streetwear outfit are taken from <Video 1> and remain unchanged in all three shots, and he is the only person who appears in the frame.
<Video 1> is the reference street-dance clip that supplies <Subject 1>'s dancing image and movement style; its original location, background dancers, and camera work are not carried over into the target video.
<Audio 1> is the voice-timbre reference for <Subject 1> (S1), supplying his mid-register male vocal timbre; only the timbre is taken from <Audio 1>, and none of its original wording or melody is carried over.
<Audio 2> is the voice-timbre reference for the off-screen female harmony singer (S2), supplying her upper-register harmony timbre; only the timbre is taken from <Audio 2>, and she never appears in the frame.
summary:
[reference generation + audio reference] A ten-second street-dance music video performed by <Subject 1> in an empty night-time basketball court. He enters on the beat, dances through the verse in the movement style of <Video 1>, and sings the lead vocal in his own timbre taken from <Audio 1>; at the chorus an off-screen female harmony voice (S2) joins on the held notes while the frame stays on him.
retention_analysis:
<Subject 1> (appears in [Shot 1], [Shot 2], [Shot 3]): fully_preserved - the dancer's face, hairstyle, build, and streetwear outfit shown in <Video 1> are retained unchanged.
<Video 1> (appears in [Shot 1], [Shot 2], [Shot 3]): partially_preserved - only the dancer's image and movement style are referenced; the original location, background dancers, and camera work of <Video 1> are not carried over.
<Audio 1> (appears in [Shot 1], [Shot 2], [Shot 3]): reference - only the lead vocal timbre of <Subject 1> (S1) is referenced, with none of the original wording or melody retained.
<Audio 2> (appears in [Shot 3]): reference - only the upper-register harmony timbre of the off-screen female harmony singer (S2) is referenced, with none of the original wording or melody retained.
detailed_description:
Live-action, cinematic, shot in a hard-edged street style with high-contrast night lighting. [Shot 1] A wide static shot of an empty outdoor basketball court at night, lit by two sodium floods on tall poles that throw long parallel shadows across the painted asphalt, with a chain-link fence catching the light along the far edge. The painted lines of the court are worn down to faint traces, and the floodlight poles stand just outside the frame so their light arrives flat and even across the asphalt. <Subject 1>, the male lead vocalist and dancer taken from <Video 1>, steps into the floods from the left, plants his feet centre frame, and begins the verse: his upper body rides the beat while his feet run a fast, low-to-the-ground footwork pattern in the movement style of <Video 1>, and he sings the lead vocal in a driving mid-register phrase, his jaw and shoulders working with the rhythm. The camera cuts to [Shot 2] At 00:03.500, a low-angle push-in that travels with him as he crosses the key: he spins once on the ball of his foot, drops into a squat and snaps back up, arms cutting the air on the off-beats, and keeps singing in the same mid-register, the sweat on his neck catching the sodium light. The chain-link fence slides past behind him as the camera travels, and the shadow under him shortens and stretches again as he crosses the pool of light. The camera cuts to [Shot 3] At 00:06.800, a slow pull-out to a medium full shot as the chorus arrives: he holds centre frame, opens his chest and sings a long sustained note while his arms sweep wide, and an off-screen female harmony voice (S2) joins him on the held note from outside the frame, stacking a bright upper register above his line, her lips never visible, since she does not appear in the frame at any point. The two voices hold together over the final bars, his footwork settling into a closed stance, and the shot ends on him standing still in the floods, his breath showing faintly in the cold air, with the empty hoop and the empty court behind him.
overall_soundscape:
Night-time street tone around the court, with a low traffic wash beyond the fence, the thin rattle of the chain-link in a light wind, and one squeak of a sneaker sole on asphalt during the footwork.
non_diegetic_music:
A hard street-rap backing track sits outside the scene as audience-only music, built on a punchy 90 BPM drum pattern with sub-bass hits and a clipped horn stab.
It drops to kick and sub alone under the sustained chorus note, then cuts out entirely on the final frame.
资产层清单
{
"references": [
{
"order": 1,
"id": "street_dance_clip",
"type": "video",
"role": "舞者形象与动作来源",
"url": "<OSS>/h3_v3_ts16/dancer.mp4",
"note": "",
"binding": "<Video 1>(<Subject 1> 的形象与动作风格来源;retention_analysis 标 partially_preserved,原场景/背景舞者/运镜不带入)"
},
{
"order": 2,
"id": "rapper_voice",
"type": "audio",
"role": "男声歌手音色来源,对应目标视频中的男主唱",
"url": "<OSS>/h3_v3_ts16/rapper_voice.mp3",
"note": "",
"binding": "<Audio 1>(音色参考,对应 <Subject 1> (S1);绑定关系内联写进 <Audio 1> 定义句,复用说话人全局 ID (S1),不独立重编号)"
},
{
"order": 3,
"id": "chorus_voice",
"type": "audio",
"role": "女声和声音色来源,对应目标视频中的女和声(不出现在画面中)",
"url": "<OSS>/h3_v3_ts16/chorus.mp3",
"note": "",
"binding": "<Audio 2>(音色参考,对应画外女和声 (S2);绑定关系内联写进 <Audio 2> 定义句,复用说话人全局 ID (S2))"
}
],
"notes": "数组顺序 = asset_menu 原始顺序 = 自动编号顺序;原 url 原样保留、未改链;url 从不进入正文层。H13:音频 2 段 + 视频 1 段同批输入,音频未单独输入(合规);合计 3 文件 ≤12。"
}
交付说明
- 形态与链路:单件需求(MV Prompt)→ 交单件 ① 型六段,未硬拆四层(§0.6 单件不硬拆)。
- 时长:
duration_seconds = 10(取自「10 秒」,不写进正文)。切镜时刻 0.000 / 3.500 / 6.800,末镜收在 10.000:H5 严格递增、毫秒 3 位、落在时长内、[Shot 1]无时间戳;正文零处窗式记号(H5b)。 <Video N>/<Audio N>独立编号(H15):<Video 1>与<Audio 1>/<Audio 2>各自独立编号,序号只表示本类内顺序、不表配对——普通参考视频<Video 1>不因含声而自动产生<Audio N>。- 音频锚定(§5):
<Audio 1>显式对应目标说话人<Subject 1> (S1)、<Audio 2>对应画外女和声(S2),对应关系内联写进<Audio N>定义句(只放资产层元数据不算完成绑定);两者都复用说话人全局 ID,不在音频定义处重新编号。 - 音色参考只取音色(doc2:577):
<Audio 1>/<Audio 2>的原歌词与旋律一律不带入,retention_analysis相应标reference;同时目标视频的歌词输入未给,故正文只描述演唱行为(mid-register phrase / sustained note),不臆造任何歌词文本——如需逐字歌词,请补原文(将按口径 V 逐字保留)。 - 女和声的画外处理:输入明文「副歌部分加入但不出现她的画面」→ 她不立为
<Subject N>(非可见内容不得立为主体),只作为发声者获得(S2),画面上写明 she never appears / lips never visible,画外 + 唇部不可见的语义照常成立。 - summary 前缀:
[reference generation + audio reference](六类型之一 ++组合,官方允许);本片不是对<Video 1>的剪辑,故不写「The target video is an edited version of<Video 1>」,<Video 1>标partially_preserved(只取形象与动作风格,场景/背景舞者/运镜不带入)。 - 关系标记(H9):仅使用 8 个固定取值中的
fully_preserved/partially_preserved/reference,无自创取值;retention_analysis逐标签一行、未写(Sx)。 - 素材认领(幽灵禁令,C 级构造性设计):3 条素材全部被认领——
<Subject 1>由 [Shot 1]–[Shot 3] 认领,<Video 1>在定义句 + 三镜的动作风格 + retention_analysis 认领,<Audio 1>贯穿三镜的演唱,<Audio 2>由 [Shot 3] 副歌认领。 - 声音分层:演唱是画面内角色能听到的同步音 → 写进
detailed_description,不重复写进overall_soundscape;overall_soundscape1 句写环境层(车流 / 铁丝网 / 球鞋摩擦);伴奏按非叙事配乐处理(观众可闻、角色听不到)→ 写进non_diegetic_music2 句(1–3 句区间内),并给出乐器、速度(90 BPM)与动态变化。 - H1 实测:最终提示词交付件 4672 字符(
len()口径)。 - H6 实测:
detailed_description363 英文词,落在 350–500。