ToonFlow provides an open-source, agent-driven pipeline that converts novels and scripts into animated short dramas. The desktop application runs locally but requires external, paid AI model services to perform text, image, and video generation.
Project overview
It combines a three-layer agent system with an infinite canvas workbench, aiming to reduce the cost and manual effort of multi-step short drama engineering from text to video.
Project type
AI Agent · Workflow · Video
Use cases
Documents & Office · Image Processing · Video Creation
Deployment
Refer to project documentation
License
Apache-2.0
Best for
Creators looking to organize scripts, characters, storyboards, assets, and video nodes in an infinite canvas for parallel production.
Users who want to extract and structure events from original chapter texts to assist with script adaptation.
Key capabilities
Organizes scripts, characters, storyboards, assets, and video nodes in an infinite canvas for free arrangement, backtracking, and parallel production.
Uses a three-layer agent system (decision, execution, supervision) that handles task decomposition, content generation, quality review, and revision feedback.
Implements a local ONNX vector retrieval-based cross-session memory system supporting short-term messages, long-term summaries, and semantic recall.
Allows users to write provider TypeScript logic directly in the settings center without changing source code or restarting.
Automatically extracts and structures events from original chapter texts to precisely call context during script adaptation.
Externalizes core prompts for ScriptAgent and ProductionAgent into Markdown Skill files for online editing and tuning.
Provides a cross-platform desktop app for content creation.
Limitations and risks
Requires external AI service interfaces, specifically large language models, Sora or Doubao video services, and the Nano Banana Pro image generation model, to function.
Video generation and other AI tasks incur costs, with an example demo cost documented at about 130 RMB.
The tool depends on external paid AI services for its core functionality.
Getting started
Download and install the application, start it and login, configure external model provider settings, and import original text to run the agents. Pre-built installer binaries and a one-command Docker Compose setup are available.
Alternatives and comparisons
An agentic video creation framework that handles scriptwriting, storyboarding, character creation, and final video generation from a concept.
A local meta-layer for coding agents that converts HTML, CSS, and data into MP4 videos without per-render fees.
A framework to build and orchestrate video agents that reason through tasks like search, editing, compilation, and generation.