Toonflow

// Overview

Toonflow is an open-source AI workstation for short-drama production. It closes the loop from planning to scriptwriting to storyboarding to final output inside one project, organising scripts, characters, storyboards, assets and video nodes on an infinite canvas rather than a linear timeline — so parts of a production can be reworked, backtracked or generated in parallel.

Its adaptation pipeline is event-graph driven. Toonflow extracts chapter events from the source novel and stores them structurally; script adaptation then pulls context through that event graph instead of re-reading raw long text, which is what keeps a long book from degrading into lost details. Three layers of agents — decision, execution and supervision — cover task decomposition, content generation, quality review and revision feedback, and a cross-session memory system backed by local ONNX vector retrieval holds short-term messages, long-term summaries and semantic recall across rounds.

Toonflow is model-agnostic and expects you to bring your own endpoints: a language model API, a video generation service, and an image generation model. Providers are programmable — you write vendor logic in TypeScript directly in the settings centre and it takes effect without editing source or restarting. The core prompts for ScriptAgent and ProductionAgent are externalised as Markdown skill files you can edit online. It ships as an Electron desktop app for Windows, macOS and Linux, and as a Docker or Node.js deployment, with a UI in seven languages.

// What it does

• Infinite-canvas workbench holding scripts, characters, storyboards, assets and video nodes in one arrangeable graph
• Three-layer agent system — decision, execution and supervision — covering decomposition, generation, review and revision
• Chapter event graph extracted from the source novel, so script adaptation calls precise context instead of re-reading long text
• Persistent cross-session agent memory built on local ONNX vector retrieval
• Programmable provider system: write vendor logic in TypeScript in the settings centre, effective instantly with no restart
• ScriptAgent and ProductionAgent prompts externalised as editable Markdown skill files

// Getting started

Toonflow needs three external services before it can generate anything: a language model API endpoint, a video generation service endpoint, and an image generation model API. Install the desktop build or deploy it yourself, then configure those vendors in the settings centre.

# https://github.com/HBAI-Ltd/Toonflow-app/releases

git clone https://github.com/HBAI-Ltd/Toonflow-app.git
cd Toonflow-app

yarn docker:local

# or build manually
docker build -t toonflow .
docker run -d -p 10588:10588 -v <local_data_path>:/app/data toonflow
# then open http://localhost:10588/index.html

nvm install 24
npm install -g yarn pm2

cd /opt
git clone https://github.com/HBAI-Ltd/Toonflow-app.git
cd Toonflow-app
yarn install
yarn build

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Adapting a long novel into a shootable short-drama script without losing chapter-level detail to context limits
• Running storyboard and video generation as an editable node graph so single shots can be re-rolled in isolation
• Prototyping video concepts against your own model endpoints and keeping project data on your own machine
• Standardising a small studio's script and production prompts as versioned Markdown skill files

// How Toonflow compares

Toonflow alongside other open-source video generation tools AI/TLDR tracks, ranked by GitHub stars.