OPEN SOURCE · TypeScript · Apache-2.0 · Video

Where it fits in the workflow
Toonflow is an open-source AI workbench aimed at short-drama and motion-comic production, built by the HBAI Tech team. It organizes the full closed loop of planning, screenwriting, storyboarding, and rendering around four stages, turning a novel or script into an animated short with one click. An infinite-canvas workbench lets you lay out script, character, storyboard, asset, and video nodes freely, with full backtracking and parallel production.
Core capabilities rely on a three-layer agent collaboration system. The first layer is ScriptAgent (the writer agent): it extracts chapter event graphs from source material and stores them in a structured form, then uses that as the basis for story scaffolding, adaptation strategy, and structured scripts — drastically reducing information loss over long texts. The second layer is ProductionAgent (the production agent): it calls text, image, and video models from the infinite canvas to generate character designs, storyboards, and shot videos. The third layer is a supervisor agent that does quality review and revision feedback to keep the final cut consistent. Every agent's core prompts are externalized as Markdown Skill files and editable from the settings panel with live effect — no source-code changes or restarts required.
A persistent agent memory is built in: a local ONNX-based vector retrieval system that handles short-term messages, long-term summaries, and semantic recall, so multi-turn creation never loses context. The programmable provider system is another highlight — write your provider logic in TypeScript right inside the settings panel and watch it take effect, which simplifies private deployment and multi-model integration. Vercel AI SDK already ships providers for OpenAI, Anthropic, Google, DeepSeek, Zhipu, MiniMax, Tongyi Qianwen, and xAI.
Deployment comes in two paths: Docker (locally yarn docker:local or docker build -t toonflow . with a mounted data volume, running on port 10588), or cloud (Node.js 24+ with PM2, then yarn install and yarn build). The Electron desktop client (yarn dev:gui) is also available as an out-of-the-box option and includes its own bundled front end. Licensed under Apache-2.0, the project currently has 11k+ stars and 2k+ forks on GitHub, plus AtomGit G-Star No.540 and Gitee GVP certifications. The UI ships in seven languages: Simplified Chinese, Traditional Chinese, English, Thai, Vietnamese, Japanese, and Russian; both Simplified Chinese and English READMEs are maintained.
Demo Videos
Screenshots

Workflow stage
Quick Start
Deployment guides and docs are linked externally to stay up to date.