integrated_multimodal_description（画面+动作+对白逐镜头）→ overall_soundscape（环境音）→ non_diegetic_music（背景音乐）。关键帧模式下先写对齐指令（如「<Picture 1> 对齐 0.00 秒」）。[Shot N] At MM:SS.mmm。运镜 = 类型+幅度+速度（如「camera pushes in with small amplitude at slow speed」），说话人用 (S1) 稳定编号，台词放 <d>[English] ...</d>。subject_definitions → summary → retention_analysis → detailed_description → overall_soundscape → non_diegetic_music。用 <Subject N> / <Picture N> / <Video N> / <Audio N> 标签管理参考素材。