rea-video

rea-video, at reavideo.ai, is an open-source skill and command-line tool for AI agents. The code is at Rea Video under the Apache-2.0 license. It runs in Claude Code, Codex, or any agent that can run commands. The README's quick start is one prompt that tells the agent to install the repo and break down a video file on your computer. It works on local files, and the site says it does not take browser uploads.

The pipeline cuts the video into shots and splits the audio into voice, music and effects. It then measures frame cadence, sensor noise, whether the whole image moves as one flat plane, and AIGC tags in the metadata. It also looks at hands, text and physics, and listens for BPM, key, and singing versus speech. Hard evidence beats how a shot looks, and conflicts are flagged for the agent to check visually. Each shot is judged on its own as live footage, AI video, avatar, 3D or animation, code or template, screen recording, or a mix.

You get a report as HTML or PDF and a VIDEO.md package that an agent can read in layers. Filmed shots come with camera positions, moves, lighting setups and color grade. AI shots come with a reverse-engineered prompt and a suggested model. Code-style shots are described so they can be rebuilt as a Remotion project. The remake plan batches shots to what the target video model can do.

Tools-only mode is offline and free. You can connect a video-capable model such as Gemini or Doubao by giving the agent its endpoint and key. The README says that makes the live, AI and 3D calls more accurate and adds a prompt for every shot, and that you pay whatever that service charges. It also says platform downloads are recompressed, which removes noise evidence and platform tags, so a vision model helps most there. The site says rea-video does not create the finished video and does not guarantee it recovers the exact original method.

Pricing: Free and open source under Apache-2.0, checked November 12, 2026 at https://www.reavideo.ai/ and https://github.com/JuneYaooo/rea-video. Tools-only mode runs offline at no cost. If you connect a video-capable model such as Gemini or Doubao, that provider bills you. Your AI assistant may also charge separately.