rea-video

# rea-video

Breaks any video down shot by shot to show how its visuals, text, editing and audio were made, then gives a plan to remake it. For creators and marketers who use Claude Code or Codex.

Made by JuneYaooo on GitHub. These notes are Mindket's, written from its README; the project's own files and license are in its repository.

## About

What it is: rea-video is a skill and command line tool for AI agents that reverse-engineers a video. It explains, one shot at a time, how the picture, the on-screen text, the cuts and the sound were produced. Then it writes a plan for making your own version, sorted into filming, generating with AI and coding.

Who it suits: It suits creators, marketers and editors who want to learn how a clip was made, or who want to remake it with their own product or story.

```table
Industry | Who uses it | How they use it | Example
Marketing | Ad teams | Break down a viral ad to see how it was built | Learn that a phone-shot TikTok ad is all live footage, with one set and two camera positions
Content creation | Short-video creators | Get a remake plan for a clip they like | See how the shots of an AI video can be batched into fewer generations
Motion design | Designers and developers | Check whether a promo is driven by code, then rebuild it | A product promo that is mostly code-driven motion graphics can be rebuilt as one Remotion project
Video editing | Editors | Pull out caption timing and styling | Export each caption line with its start and end time to SRT or ASS
```

Anyone who only wants to watch or trim a video, with no interest in how it was made, can skip it.

Where to use it: It runs inside an AI agent that can run commands, and a video model is an optional extra.

- Claude Code
- Codex
- Any other agent that can run commands
- A video-capable model such as Gemini or Doubao, which you add by giving the agent its endpoint and key

```table
Use case | Who | What they get back
Study a viral ad | A marketer | Per-shot frames, evidence, camera and lighting notes in a report.html file
Plan an AI remake | A creator | A reverse-engineered prompt for each shot, with advice on batching shots and choosing a model
Rebuild a motion graphic | A developer | The text, where it sits and when it moves, so you can recreate it as a Remotion project
Extract captions | An editor | Timing for every caption line, plus its color, outline and entrance style
```

How to set it up: You need a few things first.

- ffmpeg
- Python 3.10 or newer
- An AI agent that can run commands, such as Claude Code or Codex
- Optional: a video-capable model, with its endpoint and key

1. The simple way is to ask your agent to install the project from its GitHub page and break down a video file by its path. The agent installs the CLI and learns the workflow.
2. To install by hand, install ffmpeg, then clone the project's repository and open its folder.
3. Run the install script from inside that folder. It installs the CLI and links the skill into the agent folders.
4. Run the help command to check that the CLI works.

```bash
brew install ffmpeg            # or: sudo apt install ffmpeg
bash scripts/install_skill.sh  # installs the CLI (pip, or pipx / a local .venv if pip is blocked) and links the skill into ~/.claude, ~/.codex, ~/.agents
rea-video --help
```

How to use it: Tell your agent to break down a video and give it the file path. Reports come out in Chinese or English, whichever language you ask in, and your agent passes the --lang option for you. You get three things back.

- A one-line summary of how the video was made
- A remake plan that sorts the work into filming, AI generation and coding
- A detailed report.html page that shows each shot's frame, the evidence, a prompt, camera and lighting notes, and how the audio was built

Example: You ask your agent to break down a one-minute product promo. It replies with a summary along the lines of 41% AI video and 59% code UI, then shows the remake plan. The report lists each shot with its frame, the evidence and the prompt behind it.

For newcomers and experts: rea-video is built for AI agents, so you mostly talk to your agent and it does the work. A good first try is to ask it to break down a short clip you already have.

Each shot is judged on its own as live footage, AI video, avatar, 3D, code or screen recording. A mixed video is reassembled on its original timeline. Every video also gets notes on captions, stickers, cuts and audio, including voice type, music and BPM. A video-capable model makes the live, AI and 3D calls more accurate. Without one, an agent that can see images checks the uncertain shots itself.

```flow
You give your agent a video file
The agent installs the CLI and learns the workflow
Each shot is judged as live, AI, 3D, code or screen
It gathers notes on text, editing and audio
It writes a summary, a remake plan and report.html
```

Good to know:

- It needs ffmpeg and Python 3.10 or newer
- The long documentation is only in Chinese so far, but your agent can read it and reply in your language
- Six example reports are in English, so you can see a report without installing anything
- The sample videos in the repository are re-compressed, which strips platform tags, so your results can differ from the example reports
- The install script uses pip, or pipx or a local .venv if pip is blocked

## On GitHub, 6 November 2026

- Stars: 2,477
- Forks: 348
- Watching: 8
- Open issues and pull requests: 2
- Language: Python
- License: Apache-2.0
- Topics: agent-skills, ai-video, aigc-detection, claude-code, codex, ffmpeg, prompt-engineering, remotion, reverse-engineering, short-video, shot-detection, subtitles

## Links

- Code: https://github.com/JuneYaooo/rea-video
- Releases: https://github.com/JuneYaooo/rea-video/releases
- Issues: https://github.com/JuneYaooo/rea-video/issues
- Star history: https://www.star-history.com/#JuneYaooo/rea-video&Date

## Get the code

```
git clone https://github.com/JuneYaooo/rea-video.git
```

Then follow the setup in its README. Nothing in this kit runs by itself.

README.md  6 KB