GitHub radar
Find out exactly how any video was really made
This is a tool for your AI agent that cuts any video into shots and sound, then tells you for each shot whether it's real camera footage, AI-generated, or code-made graphics. You get an answer instead of a guess.
An AI agent splits a video into shots and audio layers, then for each shot decides how it was made: filmed with a camera, AI-generated, built with code, or a digital human with a voiceover.
When it helps
Useful when you make AI content and want to know how someone else's good video was actually made before you try to copy it. Skip it if a plain description of a video is enough.
Pros
- Runs offline and free on a regular laptop, no GPU needed
- Shows what's really filmed versus AI-generated versus code graphics
- Gives a plan to recreate the clip by shooting, AI, or code
- Recovers prompts for AI shots and counts generations needed
Cons
- Accuracy drops on downloaded, recompressed video without a paid model
- The tool itself doesn't assemble a finished video, only a plan
How to set it up — step by step
- 1Open your AI agent (Claude Code or Codex) in the folder with the video you want to analyze
- 2Send the agent the text from the block below
- 3Let the agent install what it needs on its own and break the video into shots and sound
- 4If the agent asks about a vision model like Gemini, tell it which key to use, or skip it and stay free and offline
- 5Check the one-line verdict and open the report.html with a screenshot for every shot
- 6Compare the verdict to what you expected — real footage, AI, or code-made graphics
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install rea-video from https://github.com/JuneYaooo/rea-video, then break down this video: [I'll give you a file path or link]. At the end show me the summary line and open the report.html. If you need a vision-capable model for better accuracy (Gemini, for example), ask me which key to use.
Runs on a regular laptop, no GPU needed. The basic mode without a connected model works offline and for free; the agent just needs Python 3.10+ and ffmpeg, which it installs itself. For compressed videos downloaded from platforms, accuracy is better with a vision-capable model plugged in (Gemini or Doubao) - in that case frames go to the cloud, and you pay whatever that service charges.
Open on GitHub▌ More finds