Dok & tutorial

Breakdown video pertama Anda

Panduan langkah demi langkah untuk pemula: instal rea-video di Claude Code atau Codex, analisis video lokal pertama Anda, lalu lanjut dengan direktori tutorial lengkap.

Lihat remake orang lain →

Baru di sini? Mulai dari langkah ini (sekitar 5 menit)

What you get, before you start

When the analysis finishes, rea-video writes a complete report for your video: the time range and production method of every shot (filmed, AI-generated, 3D, code graphics or screen recording), the caption and audio layers, ready-to-use per-shot prompts, and a production plan that says which shots to film, which to generate, and which assets you must prepare. The report comes as a web page (report.html) you can open in your browser and a text summary (VIDEO.md).

Before you start: three things to prepare

No editing or programming experience is needed. Get these three things ready first:

  1. Your own computerWindows or macOS both work. rea-video runs locally on your machine — your video is never uploaded to this website.
  2. An AI assistant that can run commandsSuch as Claude Code or OpenAI Codex. Think of it as a programmer living in your computer: you describe what you want in everyday language, and it downloads, installs and runs the tools for you.
  3. One reference video fileA common format such as MP4 or MOV. Use footage you filmed, exported or have permission to analyze.

You do not need to install Python or ffmpeg beforehand — in the easiest route the AI assistant installs them for you.

Step 1 · Open the “terminal”

A terminal (Terminal on macOS, PowerShell on Windows) is a window where you control your computer with typed commands. The AI assistant works here. Opening it changes nothing on your computer.

  • macOS: press Command + Space, type “Terminal” (or 终端), press Enter.
  • Windows: press the Win key, type “PowerShell”, press Enter.

A window with a blinking cursor appears — that is the terminal. Keep it open.

Step 2 · Install an AI assistant (pick one)

The easiest route for beginners is Claude Code. You need a Claude Pro/Max subscription or an API key; OpenAI Codex works similarly with a ChatGPT account.

First install Node.js (it provides the npm command used below): download the LTS version from nodejs.org and run the installer.

npm install -g @anthropic-ai/claude-code

Then type claude --version. If a version number appears, it works.

Prefer OpenAI Codex? Install with the command below, then run codex and sign in with your ChatGPT account.

npm install -g @openai/codex

Install commands can change between versions — follow the official guides at claude.com/claude-code and developers.openai.com/codex if anything differs.

Step 3 · Put your video somewhere easy to find

Move the reference video to a simple location such as your Desktop. The “path” is the file’s address on your computer — you will paste it into the instruction in the next step.

Beginner tip: drag the video file into the terminal window and its full path appears automatically. Copy it and keep it handy.

Step 4 · One sentence to install rea-video and start

Start your assistant (type claude or codex in the terminal), then paste this instruction and replace the example path with your own video path:

Instruction to paste

Install https://github.com/JuneYaooo/rea-video, then break down this video: /path/to/video.mp4

The assistant will do the rest: download rea-video, install its dependencies (including ffmpeg), analyze the video and write the report. Approve the actions it asks about, then watch it work.

The first run usually takes a few minutes; longer videos take longer. You do not need to type anything else.

Step 5 · Open the report and read three things

When it finishes, the results live in an out/<video-name>/ folder next to where the assistant ran:

  • report.html — double-click to open it in your browser. This is the full, human-readable report.
  • VIDEO.md — the text summary. Start here when you are in a hurry.

Inside the report, look at three things first:

  1. The shot table — The start time, duration and likely production method of every shot.
  2. Confidence levels — How sure the report is about each judgment — an estimate, not a guarantee.
  3. The production plan — Which shots to film, which to generate, and which assets you need to prepare.

You can also just ask the assistant follow-up questions in the same conversation, for example:

  • “How was shot 5 made?”
  • “Split the shots into ones I film and ones I generate.”
  • “Write a shot list I can take to a shoot.”

Optional: add a vision model for better judgments

rea-video works without any model configuration. Adding a vision-capable model such as Gemini or Doubao improves the visual descriptions and the per-shot prompts.

Model providers charge by usage. Keep API keys in your local configuration only, and never commit them to a repository.

Advanced: manual installation

If you prefer to install by hand, or your assistant asks for the exact steps:

git clone https://github.com/JuneYaooo/rea-video
cd rea-video
bash scripts/install_skill.sh
rea-video --help

Troubleshooting

“ffmpeg not found”
Tell your assistant: “Install ffmpeg and add it to my PATH”, or install it yourself (brew install ffmpeg on macOS, winget install ffmpeg on Windows) and run again.
Python version too old
rea-video needs Python 3.10+. Ask your assistant to check (python3 --version) and install a newer one if needed.
Model errors
Check the API key, endpoint and quota. Basic breakdowns also work without a vision model.
How long does it take?
Usually a few minutes. Longer and higher-resolution videos take more time.
Is it free?
rea-video itself is open source and free. Your AI assistant subscription or model usage is billed by the service you choose.
Is there a web version?
Not yet — rea-video runs through an AI assistant or the command line, and this website cannot analyze uploads.

Where to go next

Full project documentation ↗