Rea Video — Reverse-Engineer Any Video, Shot by Shot
Rea Video is an open-source tool that reverse-engineers any video—live-action, AI, 3D, or code—shot by shot, telling you how each frame was made and handing you a remake plan.
No breakdown yet
Rea Video Gallery
Real Breakdowns Made with Rea Video
See how creators and researchers use Rea Video to dissect videos across formats—from TikTok mixed-media clips to AI music videos and product promos.
3D Explainer · 67s · 21 shots
Product Promo · 70s · 12 shots
AI Music Video · 165s · 86 shots
Live-Action Unboxing · 72s · 29 shots
AI + Real Person Hybrid · 58s · 21 shots
What Is Rea Video?
Rea Video is a free, open-source tool built for AI agents like Claude Code and Codex. You give it a video file, and it reverse-engineers the entire production: shot classification, camera movement, lighting setup, caption timing, editing rhythm, audio breakdown, and a concrete remake recipe. It does not generate video. It generates the blueprint.
Unlike tools that only answer "is this AI or real?," Rea Video answers "how was this made, and how do I make one?" Every breakdown includes a one-line verdict, a per-shot evidence table, reverse-engineered prompts for AI shots, filming notes for live-action shots, and a production split: what to film, what to generate, and what to code.
It's built for content creators who want to learn from viral videos, MCN teams reverse-engineering competitor formats, AI video researchers studying model behavior, and editors who need to replicate a reference cut without frame-by-frame guesswork. If you search for a rea-video workflow, this is the tool that gives you a full production recipe instead of a binary label.
Start Your First BreakdownOne Video In, a Full Production Recipe Out
What Rea Video Does for You
Upload a video or point Rea Video at a file path. It splits the video into shots, classifies each one, extracts the production layers, and writes a report you can actually build from.
Shot-by-Shot Classification
Every cut is judged as live-action, AI-generated, 3D-rendered, code/template-driven, avatar, screen recording, or mixed. Ten categories, with confidence scores and evidence frames. No more "looks AI to me" hunches.
Reverse-Engineered Prompts
For AI-generated shots, Rea Video writes a prompt that would reproduce the look—camera angle, lighting, motion, style. For live-action shots, it gives you setup notes: camera position, lens choice, lighting scheme (12 presets with gear lists), and color grade direction.
Caption & Audio Deconstruction
Caption start/end times (exportable as SRT/ASS), font style, outline, animation type. Audio split into human voice vs. AI voice, library music vs. AI music, BPM, and sound-effect placement—down to which frame a whoosh hits.
Remake Plan with Cost Awareness
The report splits production into film / generate / code. It tells you how many AI generations a sequence needs (e.g., "5 dialogue shots share one set—1 generation covers all"), which model to use, and which parts you'll need to shoot or record yourself.
How It Works
Load a video.
Drop in a local file or path—TikTok, Douyin, YouTube, or anything else. Original downloads give the best accuracy.Run the breakdown.
In Claude Code, Codex, or any command-capable agent: "Install https://github.com/JuneYaooo/rea-video, then break down this video: /path/to/video.mp4." You get report.html with per-shot frames, evidence, prompts, and audio notes.Build from the recipe.
Shoot what needs shooting, generate what needs generating, code what needs coding. The companion cheap-claude-video can even generate a new version with swapped elements—same camera, motion, pacing, different content.
Rea Video by the Numbers
Open-Source Video Reverse Engineering
3,600+
GitHub Stars
500+
Forks
98%+
Shot classification accuracy on original files (tools-only mode)
Free
Apache-2.0, no API key required for core features
Rea Video vs. Generic "AI Video Detectors"
Why Rea Video Goes Further
Most AI video detectors stop at a binary label. Rea Video gives you a production plan. If you are comparing a rea-video breakdown against a generic detector, the difference is the remake plan.
| Feature | Rea Video | Generic AI Detectors |
|---|---|---|
| Output | Per-shot classification + reverse-engineered prompts + filming notes + remake plan | "Likely AI" or "Likely real" |
| AI Shots | Reverse-engineered prompt, model suggestion, generation batching | No actionable output |
| Live Shots | Camera setup, lighting scheme, lens notes, color grade | Not covered |
| Captions & Audio | SRT/ASS export, BPM, sound-effect placement | Rarely included |
| Remake Support | Full film/generate/code split with cost awareness | None |
| Cost | Free, open-source, offline-capable | Often paid, API-dependent |
What Creators Say About Rea Video
Real Feedback from the Community
Frequently Asked Questions
About Rea Video
What exactly does Rea Video do?
Rea Video reverse-engineers video. It breaks a video into shots, classifies each shot's production method (live-action, AI, 3D, code, avatar, screen recording, or mixed), extracts captions, audio, and editing data, and outputs a remake plan. It does not generate video—it gives you the blueprint to make your own.
Is Rea Video free?
Yes. Rea Video is open-source under Apache-2.0. Tools-only mode runs offline and costs nothing. You only pay if you choose to connect a third-party vision model (Gemini or Doubao) for higher accuracy on re-compressed files.
Do I need a paid API key to use it?
No. The core breakdown works offline with no API key. For platform-downloaded files that have been re-compressed, adding a vision model improves accuracy, but it's optional. You pay whatever that model provider charges, not Rea Video.
What video types can Rea Video break down?
Live-action (talking-head, tabletop, vlog, product demo), AI video (text/image-to-video, AI short drama, AI music video), 3D/animation, code-driven motion graphics, avatar/lip-sync, screen recordings, and mixed/edited videos.
How accurate is the shot classification?
On original, non-recompressed files, tools-only mode hits 98%+ accuracy across tested sets. Re-encoded platform downloads can drop to ~88% without a vision model, which is why the optional model hook exists.
Does Rea Video generate a new video for me?
No. Rea Video outputs a recipe. To generate a remake, you use the recipe with a downstream tool. The companion cheap-claude-video can take the breakdown and produce a version with swapped elements, but Rea Video itself stays focused on analysis.
What do I need to run it?
Python 3.10+, ffmpeg, and an AI agent that can run commands (Claude Code, Codex, or similar). The agent handles installation and execution from a single prompt.
Can I use it for commercial work?
Yes. The Apache-2.0 license permits commercial use. The tool is designed for content professionals who need to understand and replicate video formats for client work or original content.
Where can I find the rea-video source code and docs?
The project is hosted on GitHub at JuneYaooo/rea-video. The README covers installation, the tools-only mode, the optional vision-model hook, and the report format. Searching for rea-video on GitHub or in an AI agent prompt will point you to the same repository.
Start Reverse-Engineering Your First Video
No API key. No credit card. Just a video file and the question: how was this made?
Pricing and hosted-service availability are not included in this document. Rea Video is an open-source project; check the GitHub repository for the latest installation instructions and feature updates.