rea-video: reverse-engineer any short video shot by shot
Added NewCopyable prompt below
Published Last verified Released
Our verdict
The most useful free reverse-engineering helper we've seen for learning how a short was made: you get a per-shot HTML report and a remake plan split into what to film, generate or code, instead of guessing from the finished clip. It runs through Claude Code, Codex or any command-running agent with ffmpeg and Python 3.10+; a video-capable model (Gemini or similar) improves live/A.I./3D calls but is optional. It is not a remake button — rights and source materials are still your problem — and shot labels can be wrong on hard cases, so read the evidence before you trust a label. We have not run it ourselves.
- Who it's for
- Short-form creators, indie marketers and developers who study talking-head ads, A.I. clips, 3D explainers or kinetic type and already use a coding agent
- Leverage
- A shot-by-shot production breakdown and remake checklist from one local video file, instead of scrubbing by hand
- Effort
- Medium
- Date added
What it does
rea-video (JuneYaooo/rea-video, Apache-2.0) installs as a CLI your agent drives. It segments the file, classifies each shot (live footage, A.I. video, avatar, 3D/animation, code/template, screen recording or mixed), pulls caption timing and style, stickers, cut rhythm, voice/music/SFX notes and a remake plan, then writes report.html plus Markdown. Example reports in the repo cover mixed A.I.+UI clips, a 3D explainer, a code-driven product promo, an A.I. music video and a live unboxing. Without a vision model it still works if your agent can see images; with one it writes reverse-engineered prompts per shot.
How builders use it
- Ask your agent to install https://github.com/JuneYaooo/rea-video and break down a local MP4 you have rights to study.
- Read the one-line summary and remake plan first, then open report.html for per-shot evidence.
- When a shot is marked code/template, hand the motion notes to a Remotion or HyperFrames workflow; when it is live, treat the gear list as a checklist, not a guarantee.
- Add a video-capable model endpoint if live vs A.I. calls keep flipping on dark or static scenes.
What to expect
A bilingual-capable HTML report and remake plan for one video, with per-shot frames, labels and audio notes you can act on.
Copy this: Install + first breakdown
# Official source: github.com/JuneYaooo/rea-video (Apache-2.0). Needs ffmpeg and Python 3.10+; runs inside Claude Code, Codex or similar.
# Ask your agent:
Install https://github.com/JuneYaooo/rea-video, then break down this video: /path/to/video.mp4
# Optional: give the agent a video-capable model endpoint and key for better live/A.I./3D calls.
# Output to expect: one-line summary, remake plan, report.html with per-shot evidence.Related discoveries
See the newest picks →- ToolCreative & Media
OpenFilm: editable videos from your coding agent
MIT skill and local editor that lets Claude Code, Codex or another coding agent write a video as web pages, then lets you change any part of it by hand.
Effort: LowPrompt#Video#Open source
- WorkflowCreative & Media
Prompt videos in code with Remotion and a coding agent
Scaffold a Remotion project, add its agent skills, and describe the video to Claude Code, Codex or OpenCode.
Effort: MediumPrompt#Video#Open source
- ToolCreative & Media
HyperFrames: HTML videos from a coding agent
Apache-2.0 framework that turns HTML, CSS and seekable animations into deterministic MP4s, with agent skills for Claude Code, Codex and Cursor.
Effort: MediumPrompt#Video#Plugin / extension
- ToolCreative & Media
Animate: procedural canvas videos from Claude Code
MIT Claude Code skill that builds short animated videos as one procedural <canvas>, with story, look and storyboard check-ins before it spends time rendering to MP4.
NewEffort: MediumPrompt#Video#Plugin / extension
Get the best new A.I. picks in your inbox
The best new agent workflows and tools we’ve vetted, with a clear verdict on each. Short, occasional, useful.