TS-41 交付物
一、交付说明
- 交付形态:① 型(raw_brief 明确要「英文结构化 Prompt(六段)」)。
- 模式:全能参考(有参考素材,且素材职责是「音频复用」=结构化音频锁定)。六段顺序:subject_definitions → summary → retention_analysis → detailed_description → overall_soundscape → non_diegetic_music。全能参考模式无模式专属指令行,故不写 instruction 行。
- duration_seconds:本次取 9(请求给 9 秒,落在 H2 的 4–15 秒区间内,无需拆分)。
- 任务类型前缀:
[audio reuse](音频整段复用)。 - 素材组合(H13)显式说明:本次菜单只有 1 条音频、无任何图/视频。官方硬上限规定音频不能单独输入、必须配图或视频。本交付按「正文已完成、待调用方补 1 条图/视频素材」出稿:请调用方至少补 1 张图或 1 段视频一并提交(不改变正文六段结构,只需在 subject_definitions 增加 1 条
<Picture N>定义、retention_analysis 增加 1 行、references 数组追加 1 条记录)。若无法补料,本 Prompt 不能作为合法请求直接提交——这是 schema/素材组合层面的冲突,已显式说明,不伪造看似合规的结果。 - 台词口径(逐句):
- 「Hold the line…don't let go」与被标记听不清的第二句:口径 R。理由:音频「整段复用」,歌词的权威来源是参考音频本身;输入括注「(第二句听不清)」与素材 note 都说明这段转写是对录音的转写而非独立提供的定稿台词,命中判定顺序②(台词同时出现在参考音频资产且要求复用)。
- R 处理结果:删装饰性波浪号
~~与装饰性 emoji🎵;标点规范化为基础书面标点,两个乐句以逗号相接,感叹句在</d>前以!结束;源词与原语言不改写("Hold the line" / "don't let go" 原词保留,不翻译、不补词)。 - 听不清的第二句:按 R 写
[unclear],不猜测改写。输入 note 逐字锁定:「第二句歌词现场录音听不清,别的都清楚。」——第二句在源录音里不可辨认,故不采用任何补写词句。
- 零臆造:录音室场景细节(调音台、话筒、耳机、隔音棉)属低风险可拍摄细节,在「歌手在录音室唱副歌」这一给定事实上做可拍摄性增强;未添加任何剧情结果(无拒绝/离开/情绪结论类推断)。
- C 级标注(不得视为官方条款):本交付未使用窗式/首窗 0 秒/窗和等于总时长等 C 级构造;「音频须配图/视频、不能单独输入」的处置方式(出稿 + 说明 + 补料预案)为工程收口,官方条款本身只规定组合硬上限。
二、资产层清单(references)
| # | id | type | role(原文) | url(原样未改链) | note(原文) | 正文层落位 |
|---|---|---|---|---|---|---|
| 1 | song_ref | audio | 音频复用(歌词要逐字保留) | <OSS>/h3_v5_ts41/song.mp3 |
第二句歌词现场录音听不清,别的都清楚。 | <Audio 1>(唯一素材,无幽灵编号) |
excluded(不可用素材单列):无 —— 本次菜单内无被 note 标记为「必须替换/不要使用/带水印/错误版本」的素材,无阻塞。
素材预检结论:song_ref 可用(note 是「部分内容听不清」的事实描述,不是「必须替换/不可用」),继续产出,不报阻塞。note 已逐字进入上方清单,正文层按口径 R 落 [unclear]。
三、最终提示词(H3 Prompt,六段)
subject_definitions: is the fully reused studio recording of the chorus, carried unchanged across the entire nine seconds of the target video; its lead vocal is the voice of the female singer (S1), recorded close to the microphone with the natural room tone of the vocal booth captured along with it.
summary: [audio reuse] A female singer finishes the chorus of a song inside a professional recording studio at night; the target video keeps the studio recording as its complete audio track while the camera pushes in from a medium-wide framing of the vocal booth to a medium close-up of her at the microphone.
retention_analysis: : fully_copy - the complete studio recording, including the lead vocal, the backing track and the room tone captured with it, is reused unchanged as the target video's audio.
detailed_description: Live-action, cinematic performance footage shot inside a professional recording studio at night, with warm practical light, shallow depth of field and the still, focused feel of a take that is being kept. [Shot 1] The camera begins on a medium-wide framing of the vocal booth seen from the control room and pushes in with a slow Push In at slow speed with small amplitude, settling on a medium close-up of the female singer standing at the microphone. The room is lit by one warm desk lamp on the mixing console behind the glass, and that lamp stays the only practical light source for the whole nine seconds: its glow rakes across the console's faders and throws a soft reflection into the glass, while a diffused key from above the booth models the singer's face and the acoustic foam panels behind her fall away into shadow. She stands close to the large-diaphragm microphone with both headphones on, one hand resting lightly on the pop filter's ring, her shoulders settled and her eyes half closed as the chorus carries her through its first line. As the push in settles, she leans a fraction toward the microphone and delivers the chorus, her lips moving in exact sync with the reused vocal and the pop filter catching a small breath between the phrases. The target video's audio is the reused studio recording , which runs unchanged from the first frame to the last, so the female singer (S1) sings the chorus exactly as it was captured: [English] Hold the line, don't let go! [unclear], the unclear second line of the chorus reaching the viewer exactly as it exists in the reused audio. Behind the glass, the console's meter bridge flickers with the level of her voice, and the small red recording light above the booth door stays lit for the whole take. On the final beat she opens her eyes, lifts her chin slightly away from the microphone and lets a small, satisfied breath go as the track resolves underneath her; her hand slides off the pop filter's ring and comes to rest flat on the mic stand, and the shot holds on her face in the warm lamp light while the reused recording runs out.
overall_soundscape: The audible track is the reused studio recording , so the room is heard exactly as it was captured: a close, dry vocal over the low hum of the console and a faint bleed of the backing track inside the booth. No additional ambience is layered on top of the reused recording.
non_diegetic_music: N/A
四、机械验收记录(§1.7 十八项)
- 形态 ① 型,单件交付(raw_brief 只要一个六段 Prompt)✓
- 六段齐全且顺序正确(doc2:250)✓
- 字段名正确;全能参考无指令行,H4 不适用 ✓
- 仅
<Audio 1>+(S1),无幽灵编号;H9 音频取值fully_copy(8 取值白名单内)✓ - references 顺序 = 菜单顺序 = 自动编号顺序;原 url 未改链 ✓
<Audio 1>在 subject_definitions 定义并被 detailed_description 引用 ✓- 对白用
<d>;口径 R:删~~/🎵、标点规范化、!终止、源词原语言保留、听不清[unclear]✓ - 未引用任何不可用素材 ✓
- 无设定冲突;无未经授权剧情结果 ✓
- ① 型散文交付,无 JSON 通道要求;无残留
Shot N/S.SS占位符 ✓ - 素材总量 1/12;音频单独输入 → H13 冲突已显式说明并给补料预案,未静默提交 ✓
- 正文层编号词法仅官方四标签与
(Sx)✓ - H1:最终提示词 ≤ 7000 字符 ✓
- H2/H3:单段 9 秒在 4–15 秒内;无独立总时长指令行;未提及帧率(不写不扣分)✓
- H5/H5b:单镜
[Shot 1]无时间戳;① 型内零处窗式记号 ✓ - H6:detailed_description 为生成类,词数见下(350–500 区间内,无豁免需要)✓
- H10:overall_soundscape 2 句;non_diegetic_music 显式
N/A(配乐为画面内复用音轨,属同步音,不写进 music)✓ - H11:一句副歌落在 9 秒内不失衡;H16:本单无分镜参考图、无秒级对齐诉求 ✓
H6 词数与 H1 字符数见交付说明附注(写作时已按 len() 口径复算)。