JuneYaooo/rea-toon

What it does

rea-toon analyzes short dramas you already have on disk and produces a study of why the drama works, a standard short-drama script, and a shot-by-shot breakdown. It is aimed at people learning short-drama writing and production, and it runs inside an AI agent that can see images (Claude Code or Codex) or with a vision model exposed through an OpenAI-compatible API.

It handles live-action, AI live-action, 2D/3D animation, motion comic/meme comic, and commentary formats, adjusting what it inspects for each. A series is read with earlier episodes as context, and an interrupted series can resume where it stopped.

Each breakdown yields one index.html with three views (Study, Script, Shot script), PDF export for script and shot script, and a VIDEO.md package plus video/ assets so the agent can answer questions later without rewatching.

Quick start

The normal entry point is to tell an agent to install the repository and break down a drama with this instruction:

Install https://github.com/JuneYaooo/rea-toon, then break down this short drama: <a video file, or a folder with one video per episode>

Prerequisites are ffmpeg and Python 3.10+. The agent installs the CLI, follows the workflow documented in the skill file, and runs it to completion. Input is a local file or a folder with one video per episode, sorted by episode number in the file name; links are not downloaded. For a single file containing many episodes, the agent first confirms episode boundaries with you and splits it.

Start with about five episodes to judge output quality and token cost before processing a full series.

Constraints

Vision capability is mandatory because every shot is inspected. Without an external model, the agent answers the questions itself at no key cost, but it stops at each judgment and may take a dozen rounds per episode. A configured vision API model is billed per call and runs in parallel, which is more suitable for long series.

Token consumption is high: each episode involves dozens of frame grids, roughly 30 vision calls, so cost should be checked on a small sample first.

The generated pages and reports are currently in Chinese, since the tool targets Chinese short dramas. The project is licensed under PolyForm Noncommercial 1.0.0: personal study and research are allowed with attribution, while commercial use requires written permission per the license.

HotRepo's take

In the 2026-11-15 new-project snapshot, rea-toon recorded 102 stars and 20 forks in its first day on the Python list, a window starting 2026-11-14. That is early traction for a tool whose effective use depends on an image-capable agent and substantial token spending, so the entry cost, not the analysis concept, is the main thing to weigh before trying it.