Video Prompt Reverse

Turn a reference video into an executable, evidence-linked generation package. Distinguish semantic recreation from exact reconstruction: a text prompt can preserve scene logic, action, camera language, pacing, and style, but cannot guarantee frame-identical output. Escalate to reference frames or control signals when identity, geometry, or motion fidelity matters.

Invocation cues

Treat “视频反推提示词” and semantically similar requests as direct requests to use this skill. Common equivalents include “视频转提示词”, “从视频生成提示词”, “视频逆向提示词”, “逐镜头反推”, “分镜拆解并生成提示词”, “反推视频 prompt”, and “video-to-prompt”. The user may append a target model, such as Veo, Sora, Kling, or Wan, and may request iterative similarity optimization.

Choose the mode

• Prompt: analyze the video and produce two language-specific packages: one Chinese block containing its positive and negative prompts, followed by one English block containing its positive and negative prompts.
• Adapt: also compile model-native prompts for one or more of veo, sora, kling, wan, or generic.
• Optimize: when a callable target generator and evaluator are available, generate bounded candidate batches, score them against the reference, and revise the prompt.

If the target model is absent, use generic. If the requested fidelity is unclear, default to semantic and cinematic similarity, not identity or pixel reconstruction.

Analyze the source