TS-13.md
10.5 KB
TS-13 交付物
第 0 步判定
- 模式:4 件参考素材(场景锚点图 + 主持人立绘 + 道具特写图 + 剪辑节奏参考视频),多素材结构化锁定 → 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:12 秒,落在 4–15s 内(H2 通过)→ 4 镜切分。
- 素材总量:图片 3 + 视频 1 = 4 文件 ≤12,图片 ≤9、视频 ≤3(H13 通过),无音频单独输入问题。
- 台词:输入未给任何台词/口播文案 → 不臆造歌词或台词,主持人只做茶艺动作不开口。
最终提示词原文
subject_definitions:
<Subject 1> is the male tea host shown in <Picture 2>; his face, hairstyle, and the exact colour and cut of his mandarin-collar tea uniform are taken from <Picture 2> and remain unchanged in every shot, and he keeps his sleeves rolled back to the forearms whenever he handles the teaware.
<Subject 2> is the small Yixing zisha teapot shown in <Picture 3>, an unglazed reddish-brown purple-clay teapot whose round body, short straight spout, and domed lid knob follow <Picture 3> exactly and do not change between shots.
<Picture 1> is the teahouse interior keyframe that anchors the first frame of [Shot 1]: a timber-framed tea room with a low wooden counter, warm paper lanterns hanging in a row, and a neatly arranged tea table beside the lattice window.
<Video 1> is the reference brand film that supplies only the cutting rhythm and shot-change style of the target video; its footage, characters, locations, and original on-screen text are not carried over.
summary:
[reference generation] A twelve-second live-action guochao tea-house commercial set inside the teahouse of <Picture 1>. The host arrives at the tea table, sets down the zisha teapot, and performs a short restrained tea ritual; the cutting rhythm follows the reference film in <Video 1>, moving from an establishing wide shot to closer insertion shots of the pour, and closing on the host and the table in a calm final frame.
retention_analysis:
<Subject 1> (appears in [Shot 1], [Shot 2], [Shot 3], [Shot 4]): fully_preserved - the host's face, hairstyle, and tea uniform shown in <Picture 2> are retained unchanged in all four shots.
<Subject 2> (appears in [Shot 2], [Shot 3]): fully_preserved - the teapot's body, spout, and lid knob shown in <Picture 3> are retained.
<Picture 1> (appears in [Shot 1]): fully_preserved - the tea room layout, the low counter, the row of paper lanterns, and the lattice window are retained as the opening frame.
<Video 1> (appears in [Shot 1], [Shot 2], [Shot 3], [Shot 4]): weak_reference - only the cutting rhythm and shot-change style are referenced; no footage, character, location, or on-screen text from <Video 1> appears in the target video.
detailed_description:
Live-action, cinematic, with warm practical lighting and a restrained, editorial pacing. [Shot 1] The shot begins from <Picture 1>: a wide establishing view of the tea room, the low wooden counter on the left, the row of paper lanterns overhead, and the tea table beside the lattice window on the right. The paper lanterns swing almost imperceptibly in the draught from the doorway, and the low counter carries a stack of clean cups with a brass kettle waiting at its edge. Daylight comes through the lattice window from camera right, laying warm bars of light across the tabletop and leaving the timber frame in soft shadow. The camera dollies in at slow speed with small amplitude as <Subject 1>, the tea host in his mandarin-collar uniform, enters from the left carrying a folded white tea cloth, crosses to the table, and sets the cloth down beside a row of small cups. The camera cuts to [Shot 2] At 00:03.500, a closer two-shot over the near end of the table: the host sets <Subject 2>, the reddish-brown zisha teapot, onto the tray in front of him, turns it once so its spout faces his free hand, and lifts the kettle in his other hand. Warm light from the window now falls on the teapot's glazed shoulder while his face stays half in shadow, and the folded cloth he set down in the previous shot still lies at the near corner of the tray within reach of his left hand. The camera cuts to [Shot 3] At 00:06.500, a tight insertion on the teapot and the cup: he pours a thin, unbroken ribbon of water into the teapot, steam rising off the spout and catching the same window light, then tilts the pot and fills the small cup until its surface trembles. The stream holds steady without splashing, and a single bead of water runs down the spout and gathers on the clay before dropping into the tray. The camera cuts to [Shot 4] At 00:09.500, a slow pull-back to a calm closing frame: the host sets the teapot down, rests both hands on the edge of the table, and looks toward the cup; the lanterns behind him hold the warm background, the lattice window keeps its bars of light across the table, and the shot settles on the table, the teapot, and the filled cup, the surface of the tea going still before the video ends.
overall_soundscape:
Quiet room tone in the tea room, with the low murmur of distant conversation behind the counter and the soft creak of timber and lantern paper.
The kettle releases a thin continuous hiss while the host pours, and small porcelain clicks mark each cup being set down on the wooden table.
non_diegetic_music:
A sparse guzheng and plucked-string bed at a gentle mid tempo, entering under [Shot 1] with single notes, thickening slightly through the pour in [Shot 3], and resolving to a single sustained note over the final frame.
资产层清单
{
"references": [
{
"order": 1,
"id": "teahouse_scene",
"type": "image",
"role": "场景构图锚点",
"url": "<OSS>/h3_v3_ts13/teahouse.png",
"note": "",
"binding": "<Picture 1>(具体目标帧锚点,单独成条,由 [Shot 1] 认领)"
},
{
"order": 2,
"id": "host_portrait",
"type": "image",
"role": "主持人立绘(Subject 1 外观来源)",
"url": "<OSS>/h3_v3_ts13/host_face.png",
"note": "用 pack:portrait_real_face 版",
"binding": "<Picture 2>(只定义 <Subject 1> 外观,在 <Subject 1> 定义句内序号内联引用,不另起独立 Picture 定义条目)"
},
{
"order": 3,
"id": "teapot_prop",
"type": "image",
"role": "道具特写(Subject 2 来源)",
"url": "<OSS>/h3_v3_ts13/teapot.png",
"note": "",
"binding": "<Picture 3>(只定义 <Subject 2> 外观,在 <Subject 2> 定义句内序号内联引用,不另起独立 Picture 定义条目)"
},
{
"order": 4,
"id": "brand_cut",
"type": "video",
"role": "剪辑节奏来源",
"url": "<OSS>/h3_v3_ts13/brandcut.mp4",
"note": "",
"binding": "<Video 1>(剪辑节奏与镜头切换方式,非可见内容,retention_analysis 标 weak_reference)"
}
],
"notes": "数组顺序 = asset_menu 原始顺序 = 自动编号顺序;原 url 原样保留,未下载、未转存、未改链(raw_brief 明文「参考图地址一律用线上 OSS 原链,不要下载转存」);url 从不进入正文层。"
}
交付说明
- 形态与链路:单件需求(TVC 短片 Prompt),expected_chain 未要求分镜设计/分镜脚本 → 交单件 ① 型六段,未硬拆四层(§0.6 单件不硬拆)。
- 时长:
duration_seconds = 12(取自「12 秒」,不写进正文;正文时长只由切镜时刻体现)。4 镜时刻 0.000 / 3.500 / 6.500 / 9.500,末镜收在 12.000,H5 严格递增、毫秒 3 位、落在时长内,[Shot 1]无时间戳。 - 每份素材的落位(§2 两层模型):
- 茶馆实景图 →
<Picture 1>独立定义条目 + 首帧锚点(the shot begins from <Picture 1>); - 主持人立绘 → 输入明文「只用它定义人物长相与服装」→ 按输入约定不另起 Picture 条目,在
<Subject 1>定义句内以<Picture 2>序号内联引用; - 紫砂壶道具特写 → 输入明文为
<Subject 2>来源 → 在<Subject 2>定义句内以<Picture 3>序号内联引用,不另起条目; - 品牌短片 →
<Video 1>,只承担剪辑节奏与镜头切换方式(非可见内容),retention_analysis标weak_reference,且不写「The target video is an edited version of<Video 1>」(本片不是对<Video 1>的剪辑)。
- 茶馆实景图 →
- summary 前缀:
[reference generation]——目标视频是对参考素材的生成类新片,不是 video editing / continuation;音频无复用。 <Video 1>/<Audio N>编号:本单只有<Video 1>,无<Audio N>(普通参考视频不因含声而自动产生<Audio N>,H15 独立编号、不表配对)。- 关系标记:仅使用 8 个固定取值中的
fully_preserved/weak_reference(H9 白名单)。 - 标签五处同义(H14):
<Subject 1>(主持人)、<Subject 2>(紫砂壶)、<Picture 1>(茶馆)、<Video 1>(节奏参考)在 subject_definitions、summary、retention_analysis、detailed_description、声音段落中含义一致;retention_analysis逐标签一行、未写(Sx)。 - 台词:输入未给任何台词 → 主持人不安排开口说话,不臆造口播文案;如需口播,请补原文(将按口径 V 逐字保留)。
- 素材认领(幽灵禁令,C 级构造性设计,非官方条款):4 条素材全部被认领——
<Picture 1>→[Shot 1],<Picture 2>→<Subject 1>定义句,<Picture 3>→<Subject 2>定义句,<Video 1>→节奏 + retention_analysis。 - 声音:环境声具体到类型(人声低鸣 / 木构与灯笼纸吱呀 / 水壶嘶声 / 瓷器碰响),
overall_soundscape2 句(1–4 句区间内);配乐层观众专属,non_diegetic_music2 句(1–3 句区间内)并给出乐器、速度与动态变化。 - 运镜:四镜各自单一主导运镜(dolly in / 两 shot 过桌 / tight insertion / pull back),切镜均带来新信息(新主体动作 / 新空间关系),无矛盾运镜塞同句。
- 推定与限制(显式标注):输入未附立绘的可见特征文字,故
<Subject 1>的外观以「从<Picture 2>取得 + 跨镜不变」的绑定式写法锁定,并补可拍的结构性约束(袖口卷至前臂);若立绘实际特征与描述冲突,以<Picture 2>为准。此为低风险处理,非官方条款要求。 - H1 实测:最终提示词交付件 4970 字符(
len()口径)。 - H6 实测:
detailed_description405 英文词,落在 350–500。