TS-05.md
10.5 KB
TS-05 交付物
第 0 步:交付形态判定
- raw_brief 逐字:「帮我做条产品发布短视频:15秒,参考素材是一张我们智能手表的产品图和一段之前拍好的代言人口播视频片段,想要代言人抬手展示手表的镜头,手表外观和之前口播视频里的人物形象都要保持一致,风格科技感但不要浮夸。」
- 多素材身份 / 外观锁定(产品外观 + 人物形象锚定)+ 结构化绑定关系 → 全能参考模式,六段交付(§3.3)。
- 需求只给一个镜头意图(「代言人抬手展示手表的镜头」),未要求分镜脚本 / 分镜设计文档、未要求按时间窗切分显示内容 → 非 ② 型,交付 ① 型(§0.2 第 2/3 条);正文英文(doc2:242)。
- 素材预检(§1.2):asset_menu 两条素材 note 均为空字符串 → 无禁止 / 替换 / 版本要求,无不可用素材,继续产出,不报阻塞;数量总闸 H13:图 1 ≤ 9、视频 1 ≤ 3、音频 0、合计 2 ≤ 12,组合合法(无单独输入的音频)。
- 时长:用户给定 15 秒,落在 4–15s 上界内(H2 满足),无需拆分方案;
duration_seconds = 15不写进正文,正文时长只由[Shot N]切镜时刻体现。
最终提示词原文
subject_definitions:
<Subject 1> is the male spokesperson shown in <Video 1>; his facial features, hairstyle, skin tone, and outfit are taken from <Video 1> and stay identical to that footage, with no change to his face, his hair, or his clothing.
<Subject 2> is the smartwatch established in <Picture 1>; its case shape, dial layout, crown and button placement, strap design, and the position of the brand mark on the dial are reproduced exactly as in <Picture 1>, and the dial and the brand mark stay clearly readable whenever his wrist is raised toward the camera.
<Video 1> is the identity anchor for <Subject 1>, providing his face, hairstyle, and clothing as they appear in the original footage; the original room, framing, and pacing of <Video 1> are not carried into the target video.
summary:
[reference generation] A 15-second product-release spot in a clean, cool-toned live-action commercial style with restrained, non-flashy production design. The spokesperson from <Video 1> stands in a dark studio and raises his wrist to present the smartwatch defined by <Picture 1>, first in a medium shot and then in a close-up that lets the case, dial, and strap read clearly. No dialogue is spoken in the target video.
retention_analysis:
<Subject 1>: fully_preserved - the spokesperson's facial features, hairstyle, and outfit from <Video 1> are retained without change.
<Subject 2>: fully_preserved - the smartwatch's case shape, dial layout, crown and button placement, strap design, and brand mark position from <Picture 1> are retained without change.
<Picture 1>: fully_preserved - the watch's front three-quarter appearance, including the position of the brand mark on the dial, is retained as the authoritative appearance source for <Subject 2>.
<Video 1>: weak_reference - only the spokesperson's face, hairstyle, and clothing are used to define <Subject 1>; the original room, framing, and pacing of <Video 1> are not carried into the target video.
detailed_description:
Live-action commercial footage with a clean, cool-toned look and restrained production design: one soft key light from the upper left, a dark graduated backdrop, and no busy set decoration, so the watch and the spokesperson carry the frame and the floor falls away into shadow. [Shot 1] The spokesperson shown in <Video 1> stands centered in the frame against the dark backdrop, the key light laying a soft edge along his shoulder and jaw while a low fill keeps his features readable and a faint rim light separates him from the background. His arms rest at his sides, and the watch sits on his left wrist under the cuff, the strap barely visible at the edge of his sleeve. He shifts his weight once, then lifts his left arm in one smooth motion, turning his forearm so the watch face swings up into the light, and the camera begins a slow Push In at slow speed from the medium shot toward his wrist; the strap flexes with the turn, the case catches a narrow highlight that travels across its edge, and the brand mark on the dial comes up facing the camera. He holds the raised position with a small, natural adjustment of the fingers, his shoulders relaxed and his gaze staying on the watch, and the sleeve falls back from the strap so the case and the first links of the strap read clearly against the dark backdrop. The shot cuts to [Shot 2] At 00:07.000, a close-up on the raised wrist against the same dark backdrop: the camera holds a Static Shot while the light does the work, a narrow highlight crossing the case and the dial as his forearm rotates a few degrees one way and back, the strap texture and the stitching reading clearly, the crown and the side button catching a short glint, and the brand mark staying legible in the center of the dial. The rim of the case throws a thin line of light along his wrist, and the dark backdrop holds soft and even with no detail to pull the eye away. He settles the wrist into its final angle, the highlight coming to rest along the case edge, and the shot ends with the dial facing the camera, the watch sharp and the reflection of the key light sitting still in the crystal.
overall_soundscape:
A quiet studio room tone runs under the whole spot, with a very low hum from the lighting rig and no street noise. A soft fabric rustle accompanies the lift of his arm and the turn of the forearm, and a single short, precise watch click is placed as the wrist settles into its final angle. No dialogue is spoken, and no footsteps are heard.
non_diegetic_music:
A restrained modern electronic underscore around 92 BPM, carried by a warm synth pad over a soft, clean pulse, with one gentle lift placed as the arm rises and a second, softer lift on the cut to the close-up. The mix stays low and even, with no drops, no stingers, and no heavy low-end hits.
资产层清单
asset_menu 已随输入送达,数组顺序 = 菜单原始顺序 = 模型自动编号顺序;原 url 原样保留、不改链;url 不进正文层。
references:
[
{
"order": 1,
"id": "watch",
"type": "image",
"role": "产品外观来源",
"note": "",
"url": "<OSS>/h3_v3_ts05/watch.png",
"落位": "以 <Picture 1> 序号内联引用于 <Subject 2> 定义句(仅定义外观,不另起独立 Picture 定义条目,§2 落位表);retention_analysis 单列一行",
"可用性": "可用(note 为空,无禁止 / 替换要求)"
},
{
"order": 2,
"id": "speaker_video",
"type": "video",
"role": "人物形象与身份锚定",
"note": "",
"url": "<OSS>/h3_v3_ts05/speaker.mp4",
"落位": "独立 <Video 1> 定义条目,为 <Subject 1> 提供身份锚定;retention_analysis 标 weak_reference",
"可用性": "可用(note 为空,无禁止 / 替换要求)"
}
]
交付说明
- 「一张智能手表的产品图」+「人物形象都要保持一致」→ 全能参考六段:手表与代言人各立一个 Subject,
retention_analysis均标fully_preserved(可见内容 4 取值之一);视频只取人物身份,原始场景 / 景别 / 节奏不带入 →<Video 1>标weak_reference(H9 八个固定取值内,无自创取值)。 - 「手表外观…保持一致」→
<Subject 2>以<Picture 1>序号内联锁定:产品图只提供外观、非首帧锚点(首帧是代言人站立画面,与「抬手展示」一致),故按 §2 落位表不另起独立<Picture 1>定义条目,而以序号内联引用完成认领;retention_analysis仍单列<Picture 1>一行以保持标签同义(H14 五处同义)。 - 代言人具体特征未在正文自创:输入未给发色 / 服装等可文字化特征,正文以
<Video 1>承担身份锚定并声明「与素材一致、不改脸 / 发 / 服装」,避免编造与真实素材冲突的外貌(§1.4 零臆造优先于「人物类 Subject 必须给 2-3 条特征」的字面满足——该条款在人物特征唯一正源是参考视频、且输入未给特征清单时的执行方式是由素材承担,不在正文虚构)。若后续拿到代言人特征清单(发色 / 发长 / 服装颜色),可直接补进<Subject 1>定义句,其余不动。 <Video 1>含口播但未设<Audio 1>(H15):普通参考视频不因含声而自动产生<Audio N>;asset_menu role 是「人物形象与身份锚定」非音色参考,且需求未要求新口播 → 不立音频条目、不写台词。目标视频无发声事件 → 全片无(Sx)(从不发声的角色不获得说话人 ID,H12/§5),V/R 口径本单不适用。- 「风格科技感但不要浮夸」→ 元指令转译(§1.4):落为具体选择——单一柔主光从左上、深色渐变背景、窄条高光扫过表壳、配乐 92 BPM 温和合成器 pad 且无 drop / 无 stinger;「科技感」「不浮夸」字样不进正文,8K / 电影感类空词亦未堆砌(§6 空词黑名单)。
- 时长与切分:
duration_seconds本次取值 = 15(用户给定,H2 上界内,不写进正文);两镜[Shot 1]0–7.0s 与[Shot 2] At 00:07.000–15.0s,时间戳毫秒 3 位、严格递增、落在时长内、无缝隙无重叠;首镜无时间戳(H5);零处窗式记号(H5b);两镜为「中景抬手 → 特写表盘」,切镜带来主体尺度新信息,非距离微调。 - 无台词 → 不新增口播内容:raw_brief 未给任何台词,不代拟代言人文案;
overall_soundscape显式写明「No dialogue is spoken」,声音四项齐(对白缺省声明 / 环境声 / 配乐 / 音效位置)。 overall_soundscape3 句(1–4)、non_diegetic_music2 句(1–3)(H10)。- C 级构造登记:本单不涉及窗式边界、链路四层等 C 级项;「非锚点图以 Subject 定义句内联引用认领」出处为 §2 落位表(官方条款 doc2:398/:400-405 配合 R 级镜像收口),非本技能自创官方条款。
- 机械验收(§1.7):六段顺序正确(doc2:250);
summary带方括号任务类型前缀[reference generation](六类型之一,非编辑源视频故未加The target video is an edited version of <Video 1>.);retention_analysis逐标签一行、沿用subject_definitions含义、零处(Sx);<Video 1>/<Picture 1>编号一致、无幽灵编号、无幽灵素材;无官方外编号词法(无object N/background N/ 小写picture n);正文 ≤ 7000 字符。