Key takeaways
- Find out how a video was made: for each shot, was it filmed, AI-generated or built in code?
- Get a remake plan that splits the job into what to film, what to generate and what to code.
- Receive a full HTML report with per-shot frames, evidence, prompts, and camera and lighting notes.
- Tools-only mode is offline and free. A video-capable model is optional, and you pay that service's charges.
- The catches: it does not render video, platform downloads need a visual check, and reusing others' footage is off limits.
What it is
rea-video describes itself as a way to reverse-engineer any video, shot by shot, saying whether it was filmed, AI-generated or coded, and how to remake it. It is open source under the Apache 2.0 licence, and the code is on GitHub at JuneYaooo/rea-video.
The problem it solves
Anyone who wants to learn from a good short video, or recreate its look on a budget, has to guess how it was made. Was that a real camera, a video model or motion graphics?
The project is built for AI agents and does not render a video itself. It works out how each shot was made, so you can decide what to shoot, generate or code. Its stated aim is learning a video's structure, pacing and techniques.
What you get
- A remake plan: the work is split into what to film, what to generate and what to code.
- A full HTML report with per-shot frames, evidence, prompts, and camera and lighting notes, plus the audio breakdown.
- A free route: tools-only mode is offline and free, according to the project.
- Flexibility: a video-capable model is optional, and without one an agent that can see images checks the uncertain shots.
- Partial automation: you film the live shots, and the agent writes the code shots as a Remotion project.
The catches
- Copying: remaking other people's videos is a legal and ethical minefield. The project says not to reuse other people's footage, music, characters or likeness, and to follow platform rules and local AI-content labelling requirements when you publish.
- Quality: platform downloads are recompressed, which wipes out noise evidence, so the project says a visual check is needed. Its accuracy figures are its own; we have not reproduced them.
- Cost: a video-capable model means paying whatever that service charges, and automatic generation only calls paid models after you confirm.
- Requirements: it needs ffmpeg and Python, and it is built to run inside an AI agent.
- Maturity: the documentation is in Chinese for now.
What we did not test
We did not install rea-video, analyse a video or try a remake. We did not check its accuracy claims or the quality of its remake plans. Every claim here comes from a literal quote in the project's README or the GitHub API, and the feature descriptions are the project's own.
Verdict
Editors' view: this suits video makers and AI-video experimenters who want to understand a technique rather than guess at it, and who already work with an AI coding agent. If you only want a quick lookalike, it is not a renderer. Treat the output as study notes, and keep your remakes original.
Sources
Need help adopting this, or have a software project in mind?
Tell us what you are building. The initial consultation is free. Email [email protected] or follow @zolnaltd.
Start a conversation →