Open source · Apache-2.0 · JuneYaooo

rea-video

Reverse-engineer any video. Shot by shot: was it filmed, AI-generated or coded — and how to remake it.

rea-video is an open-source Python project by JuneYaooo. It is built for AI agents (Claude Code, Codex, or any agent that can run commands): install the CLI, break down a video, and get evidence-backed remake guidance.

View on GitHub Star the repo Try a hosted video to prompt tool on this site

The link above is this site’s hosted video to prompt tool — not the official app of the JuneYaooo/rea-video repository.

What rea-video gives you

From the project README, a typical run produces:

One-line summary
e.g. mix of AI video vs code UI vs real screen recording, music BPM
Remake plan
Split into what to film, what to generate, and what to code
Full report.html
Per-shot frames, evidence, prompts, camera and lighting notes, and audio breakdown

Video types it analyzes

Type	Examples	What you get
Live footage	Talking-head, tabletop, handheld product demo, vlog	Setups, camera positions, framing, moves, lighting, grade
AI video	Text/image-to-video, AI short drama, AI music video	Evidence, reverse-engineered prompt per shot, generation batching
Avatar	Audio-driven portrait, lip-sync	Which shots are avatar; portrait & audio needs
3D / animation	Blender/C4D, product animation, 2D, stop-motion	Video-model vs 3D-code path
Code / template	Kinetic type, UI demos, charts, slide cards	Text, layout, motion timing (e.g. Remotion-oriented)
Screen recording	Tutorials, web demos	Which parts you must re-record
Mixed / edits	AI + code UI + real capture; live + AI edits	Per-shot judgment, reassembled on the original timeline

Also covered: captions & styled text (SRT/ASS), stickers/emoji/PiP, cut rhythm, audio (human vs AI voice, library vs AI music & BPM, SFX), and materials you must supply yourself.

Quick start

rea-video is built for AI agents. In Claude Code, Codex, or any agent that can run commands, you can say:

Install https://github.com/JuneYaooo/rea-video, then break down this video: /path/to/video.mp4

Requirements: ffmpeg and Python 3.10+.

Install by hand (from README):

brew install ffmpeg   # or: sudo apt install ffmpeg
git clone https://github.com/JuneYaooo/rea-video && cd rea-video
bash scripts/install_skill.sh
rea-video --help

Optional: hook a video-capable model via endpoint + key (see the project README) for more accurate live/AI/3D calls and reverse prompts per shot. It also works without; agents that can see images can review uncertain shots.

How it works

1
Split
Cut the video into shots; split audio into voice, music, and effects.
2
Gather evidence
Measure (frame cadence, sensor noise, flat-plane motion, AIGC metadata tags), look (hands, text, physics), listen (BPM, key, singing vs speech).
3
Decide
Hard evidence beats “looks like”; conflicts flagged for agent visual check.
4
Plan the remake
Batch shots for what the target video model can do (clip length, multi-shot, reference images).

rea-video does not render a finished remake video by itself; it explains how the original was made and how to remake it. Automatic remake of AI shots can pair with capsule-cinema (same author).

Accuracy

README reports how-it-was-made accuracy weighted by shot duration: tools-only is very high on original files; recompressed platform downloads need a vision model and/or agent visual review to stay above ~99%. Real-asset detection ~97–99%; shot boundaries reported as all correct. Reproduce via eval/RESULTS.md in the repo.

rea-video FAQ

What is rea-video?

An open-source project by JuneYaooo that reverse-engineers videos shot by shot — filmed, AI-generated, or coded — and explains how to remake them. Repo: github.com/JuneYaooo/rea-video

Is this page the official project site?

This page introduces the GitHub project. The source of truth is always the repository (license, docs, releases).

Does it need a paid vision model?

No. Tools-only mode is offline and free and already accurate on original files. For recompressed downloads, README suggests adding a video-capable model or an image-capable agent.

Can it remake the whole video automatically?

Partly. With capsule-cinema, AI shots can be generated and placed back on the timeline; you film live shots; code shots can become a Remotion project. See the repo’s docs/replicate.md.

How is this different from one-click remake tools?

Per README vs Hypit: those tools generate a lookalike; rea-video focuses on how it was made (cameras, lighting, per-shot prompts/plans) and does not render the final video itself.

Can I just repost someone else’s video?

No. README: use it to learn structure and remake with material you have rights to; follow platform rules and AI-content labeling.

Where is community support?

GitHub Discussions, Discord (link on README), and X @yao_june12106.

Credits

README: inspired by REA (Reverse Engineer Anything) — work back from what you see to how it was made, back claims with evidence, verify by measuring.

Explore the project

GitHub: github.com/JuneYaooo/rea-video
Docs in repo: docs/what-it-detects.md, docs/replicate.md, docs/cases.md, docs/cli.md

View on GitHub Hosted video to prompt on this site

Separate product experience on this site — not the official app of the JuneYaooo/rea-video repository.