AI Video Prompt Reverse Master

Problem

When replicating a video, the hard part is not noticing what makes it work, but translating camera movement, lighting shifts, color balance, BGM frequency, and transition rhythm into usable prompts. Without that translation, the result often collapses into vague phrases like “cinematic” or “premium atmosphere.”

How It Works

The skill accepts video or stills as input. It first checks FFmpeg availability, then extracts keyframes, metadata, and audio. It analyzes each frame for subject, composition, lighting, color, and material texture, with confidence labels. For adjacent shots, it identifies hard cuts, fades, dissolves, wipes, match cuts, and motion transitions. For videos with more than three shots, it outputs a shot timeline and five-element annotations. It then generates platform-specific prompts for Midjourney, Sora, Kling, Jimeng, and Doubao, plus BGM suggestions. It also supports an image-to-video workflow: the prompt avoids restating character appearance from the reference image and focuses on camera movement, action, lighting, and mood. Copyright scanning, distortion prevention, and consistency constraints help reduce face drift, costume changes, background flicker, and action breaks.

Limits

It is intended for reverse-engineering prompts from existing video or images, not for general video editing. It cannot guarantee accurate song identification, and missing FFmpeg disables audio analysis. For complex characters, fast motion, or multi-shot consistency, manual review is still required.