TS-14.md 10.7 KB

TS-14 交付物

1. 最终提示词原文(① 型多镜 / T2VA,英文正文 + 台词原语言)

{
  "integrated_multimodal_description": "Live-action period drama with a warm, painterly lantern-lit look and shallow night-market depth of field, unfolding over nine seconds. [Shot 1] A static shot with a small-amplitude push in at slow speed opens on a Ming-dynasty night market lane, where the young scholar Li Mubai (S1) stands before a lantern stall on the left of the frame. He is a slender man in his twenties in a dark indigo scholar robe with a pale grey sash, his long black hair tied up beneath a plain dark cloth cap. Two tiers of round red and orange paper lanterns hang above the stall, and at his eye level rests a rabbit-shaped lantern with white paper skin, thin bamboo ribs and a small red dot for each eye. The lantern glow is the dominant light source, warming the left side of his face while the right side stays in soft shadow, and the same glow brushes the wooden counter and the stacked unused lanterns behind him. He tilts his head and settles his gaze on the rabbit lantern, then lifts his right hand halfway and stops short of the paper skin. The camera cuts to [Shot 2] At 00:03.000, a tracking shot at slow speed that moves with him as he turns from the stall and walks toward the stone bridge, where the flower girl Ahe (S2) stands holding a shallow wicker basket. She is about sixteen, with a single braid over her shoulder, a faded rose-red jacket over a plain white inner robe, and a light blue apron tied at the waist. She lifts one branch of white plum blossoms out of the basket and offers it to him with both hands, and he reaches out and takes it. The bridge lantern behind her keeps her face readable as he asks: the man (S1) says: <d>[Chinese] 借问姑娘,这梅几钱?</d> She shakes her head slightly and presses the branch back into his hand: the girl (S2) says: <d>[Chinese] 郎君说笑了,拿去吧。</d> The camera cuts to [Shot 3] At 00:06.500, a tracking shot at slow speed that follows the two of them from behind at shoulder height as they walk side by side away from the bridge and deeper into the market, the plum branch held loosely in his right hand between them. Their figures grow smaller against the layered lantern light, and the shot holds on their backs until the lane closes around them.",
  "overall_soundscape": "Continuous night-market ambience runs under the whole clip: a low crowd murmur from beyond the lantern stalls, distant vendor calls, and the faint creak of the wooden stall frame. Footsteps on packed earth, the rustle of robe sleeves and the soft hiss of the candle flames inside the paper lanterns sit close to the two characters. The ambience thins as they walk away from the bridge, leaving their footsteps and the rustle of the plum branch foremost.",
  "non_diegetic_music": "A sparse score for solo plucked strings and a soft bamboo flute enters quietly under the first shot, keeps a slow harmonic rhythm, and swells gently as the two walk away together before thinning to one sustained note."
}

2. 资产层清单

无。raw_brief 未给任何图 / 视频 / 音频素材(无 asset_menu 字段),故无 references 数组、无 <Picture N> / <Video N> / <Audio N> 条目,模式判为 T2VA(纯文本原创场景)。素材总量 0,H13 上限不触及。

3. 交付说明

  • 交付形态 = ① 型多镜(非 ② 型)。判据:raw_brief 要求「麻烦拆成三个镜头」并给定连续故事情节,但没有任何逐窗显示内容要求(无「每隔几秒显示什么」「每窗出现什么 UI」),也没有要求交付分镜脚本 / 分镜设计文档。按 §0.2 第 3 条,「拆成 N 个镜头」本身不构成 ② 型判据——两者都沾时才判 ②,本单只沾「镜头切分」一边,故走 ① 型 [Shot N] + At MM:SS.mmm。因无分镜脚本交付物要求,本单为单件交付,未拆链路四层(§0.6「单件需求不要硬拆四层」)。
  • 模式 = T2VA:无参考素材 → 五模式决策树第一分支。T2VA 无指令行,直接从三字段开始;三字段顺序 integrated_multimodal_description → overall_soundscape → non_diegetic_music。因非全能参考模式,无 subject_definitions / summary / retention_analysis,故 H9 关系标记、H14 五处同义、H15 独立编号在本单无落点;§1.7 第 21 项的 summary 任务类型标记亦随 summary 字段一同不适用。
  • 语言:① 型正文英文;两句台词在 <d> 内保留原语言(中文)与原标点(, ? 。),逐字未改写、未补删标点、未翻译。可见文字不变量无落点(正文未自创任何招牌 / 字幕文字)。
  • 台词口径 = V(两句均适用):判定顺序第 ③ 步——输入显式声明台词为「我给的原文,一字不要动」,且无参考音频资产,故两句都按口径 V 逐字逐标点保真。未出现画外音,未用 says in an off-screen voiceover;无跨镜台词,未用 <scenetrans> / <cutoff>。
  • 说话人编号:按实际发声事件顺序分配 (S1) = 李慕白(先发声)、(S2) = 阿禾;两人在 [Shot 2] 内各说一句,编号跨镜不重号不交换。无从不发声的第三角色,故无第三个编号。
  • 时长:duration_seconds 取 9。三镜切分 0–3.0 / 3.0–6.5 / 6.5–9.0 秒,合计 9.0 秒,落在 H2 的 4–15 秒内;正文未写独立的「总时长 X 秒」指令行,时长信息由首句风格声明内的自然短语(unfolding over nine seconds)与切镜时刻共同承载。
  • 台词与镜长对齐(H11):两句台词共约 18 个音节,落在 [Shot 2] 的 3.5 秒内,一问一答各约 1.5 秒,无「短镜说长段」失衡。
  • 运镜:每镜单一主导运镜([Shot 1] Static Shot + 小幅慢速 push in;[Shot 2] / [Shot 3] Tracking Shot 慢速),无矛盾运镜塞进同一句。切镜均带来新信息(新空间:灯摊 → 桥 → 集市深处;新主体:阿禾入镜;新状态:并肩远去),三处均用 The camera cuts to,未用 cross-dissolve / fade / wipe。
  • 情节边界(§1.4):情节走向完全照 raw_brief,未改动、未增删事件;阿禾的反应用「shake her head slightly / press the branch back into his hand」这类中性短暂动作承接,未替用户添加拒绝 / 依恋 / 离开等情绪结论;「两人并肩走向集市深处」是用户给定结局,照写。
  • 低风险推定(显式标注,§8.1 CL4):① 情绪基调按「明代夜市 + 递梅」推定为温润舒缓;② non_diegetic_music 补了一段古琴 + 竹笛的克制配乐(用户未给 BGM 要求,未推定为静音,故不写 N/A)。两者均不改变主体事实。人物外观细节(李慕白的深蓝学袍 / 灰腰带 / 布帽、阿禾的辫子 / 玫红袄 / 蓝围裙)属于「人物类 Subject 需 2-3 条可辨识特征」的必要具体化,非臆造新主体——主体身份与姓名均来自 raw_brief。
  • 光源唯一性(§1.6 第 1 条):正文明确「The lantern glow is the dominant light source」,未引入第二光源;月亮 / 桥灯仅作环境光描述,不写成与灯摊光竞争的主光。
  • C 级构造标注:本单未使用任何 C 级构造(无窗边界、无幽灵禁令落点、无链路四层、无 master 序号回指),以上形态与模式判定依据均为 SKILL 内部规范或官方条款,未把内部规范表述为官方规则。

4. 交付前机械验收(§1.7,21 项逐项实测)

# 项 实测 结果
1 交付形态判定 ① 型多镜(§0.2 第 3 条),单件交付,无链路 ✓
2 段落 / 字段齐全且顺序正确 T2VA 三字段,顺序 integrated_multimodal_description → overall_soundscape → non_diegetic_music ✓
3 字段名完全正确 3 个字段名逐字对照 §3.2,无拼写错、无自造字段;T2VA 无指令行,H4 不适用 ✓
4 标签定义 / 编号一致 / 无幽灵 正文层 0 个 <Subject N> / <Picture N> / <Video N> / <Audio N>(无素材),无幽灵编号;仅 (S1) (S2) 两族记号 ✓
5 references 顺序 = 菜单顺序 无素材 → 无 references,空集一致 ✓
6 非锚点图被认领 无素材,无认领义务 ✓
7 <d> 包裹 + 逐字保真 <d> 共 2 处,每处只含语言标签 + 台词;grep 复算 2 条:<d>[Chinese] 借问姑娘,这梅几钱?</d>、<d>[Chinese] 郎君说笑了,拿去吧。</d>,与输入逐字含标点一致;口径 V,无 [unclear] ✓
8 未引用不可用素材 输入无 note,无不可用素材 ✓
9 无设定冲突 / 无未经授权剧情结果 情节三段照输入;新增仅中性动作与低风险细节 ✓
10 JSON 可解析、无 schema 外字段、无占位符 json.load 通过;字段恰 3 个;grep Shot N / S.SS = 0 处 ✓
11 H13 素材总量与组合 0 素材,组合合法 ✓
12 正文层数量与词法 grep object N / background N / 小写 picture n = 0 处;官方四标签 0 处;(Sx) 4 处(S1×2、S2×2) ✓
13 H1 ≤ 7000 字符 len() 实测 2939 字符(含标点与空白) ✓
14 H2/H3 切镜 0.0 / 3.000 / 6.500,末镜至 9.0 秒,在 4–15s 内;无独立总时长指令行;全篇未提及帧率(H3 不写帧率不扣分) ✓
15 H5/H5b [Shot 1] 无时间戳;At MM:SS.mmm 实测 2 处(At 00:03.000、At 00:06.500),毫秒均 3 位、严格递增、落在 9s 内;窗式记号 [N 秒-M 秒] 实测 0 处 ✓
16 H6 detailed_description 英文词数实测 403 词(正则 [A-Za-z0-9'-]+ 计数),落在 350–500;无豁免需要 ✓
17 H10 overall_soundscape 实测 3 句(1–4 内);non_diegetic_music 实测 1 句(1–3 内);环境声非 N/A(用户未要求静音) ✓
18 H11/H16 台词 18 音节 / 3.5s 镜长不失衡;无分镜参考图,无映射义务;无幽灵素材 ✓
19 自查声明逐项实测 上表所有数字均由脚本实测后写入:字符 2939、词 403、句 3 / 1、At 记号 2、<d> 2、(Sx) 4、[Shot N] 3、窗式记号 0、占位符 0;资产层 JSON 无 schema 外字段(本单无 JSON 资产记录);URI 字段只放 url(本单无 URI 字段) ✓
20 retention_analysis 零 (Sx) 无 retention_analysis 字段;正文内 (Sx) 4 处全部位于 integrated_multimodal_description,未落入任何关系标记行 ✓
21 summary 任务类型标记与输入对齐 无 summary 字段(T2VA 三字段结构);未标 keyframe completion(无帧锚点输入) ✓