# V18 Director prompt 结构模板（REF2VA 模式） ## 完整六段结构（英文，供 H3 使用） subject_definitions: <Subject 1> is the ... in <Picture 1>, with ... <Picture 1> is the ... reference sheet and identity anchor for <Subject 1>; ... <Subject 2> is the ... in <Picture 2>, ... <Picture 2> is the opening-frame scene anchor for [Shot 1] ... ... summary: [reference generation] A <N>-second single-take clip with no cut. <Subject 1> from <Picture 1> ... (动作/情节概要，含机位、运镜、人物走位) retention_analysis: <Subject 1> (appears in [Shot 1]): fully_preserved - ... <Picture 1> ([Shot 1] identity): fully_preserved - ... <Subject 2> (appears in [Shot 1]): fully_preserved - ... ... detailed_description: [Shot 1] <风格描述>. Vertical 9:16. One unbroken shot, no cut, for the full <N> seconds. <时间段>：...（逐段描述画面、运镜、人物动作、表情、台词） Forbidden: ... Camera: ... overall_soundscape: <环境音效描述>（脚步声、环境声、人声台词位置） non_diegetic_music: <背景音乐风格描述> ## 规则 - <Picture 1>、<Picture 2>... 对应上传的参考图（timeline items 的 slot 0,1,...） - <Subject N> 为主语定义，必须引用对应 <Picture N> - retention_analysis 用 fully_preserved / weak_reference / not_referenced 三级 - detailed_description 要包含机位、运镜、时间轴分段