Toonflow is an MIT-licensed, open-source AI creative platform for short-drama, comic-drama and short-video production that unites scriptwriting, asset management and image and video generation on a single infinite canvas, made for creators and teams who want that workflow running on their own hardware rather than inside a closed hosted service.

What it is

Toonflow is a TypeScript application that combines an infinite canvas, AI agents and visual workflows for image generation, video generation and intelligent storyboarding. It belongs to the generative-AI creative-tooling ecosystem alongside node-based editors, local ComfyUI pipelines and LLM backends, and it is distributed three ways: as a cross-platform desktop client, as a Docker deployment, and as a Web plus Server install for Linux hosts. Projects, configuration, plugins and assets are kept on the user's own device or server rather than in a vendor's cloud.

The concrete problem it solves is fragmentation. Script, character, scene and prop assets, storyboards and generated clips normally live in separate documents, folders and generation tools, and moving between them breaks the creative thread. Toonflow replaces that patchwork by organising script, characters, scenes and video clips inside one infinite canvas, and by keeping the model layer swappable — third-party APIs, or a locally run ComfyUI and LLM. The project describes its canvas-based approach as similar to LibTV and TapNow, with the difference that it is open source and self-hostable.

Key capabilities

• An infinite canvas that organises scripts, characters, scenes and video clips in one workspace, with node menus and node grouping for arranging the flow.
• Free model connection: third-party APIs can be configured, and a local ComfyUI instance and LLM can be attached instead.
• MCP support for linking external tools and services to the platform.
• A plugin marketplace at api.toonflow.net/console/plugIn that extends nodes, tools and creative capabilities.
• An open Agent layer that exposes prompts, tools and A2A, so agent behaviour can be customised and agents can collaborate with external systems.
• Local deployment across desktop, Docker and server, so projects and source assets remain on hardware the user controls.
• Production aids including a 3D director's stage with shot previews, character three-view sheets and multi-reference-material video generation.

Who uses it and how

• Creators and teams making short dramas, comic dramas and short videos, who keep script, cast, scene and prop assets beside the generated clips instead of spread across separate applications.
• Desktop users on Windows and macOS (Apple Silicon), installing a release build; the Windows installer detects and installs WebView2 automatically.
• Self-hosters using the Docker path: clone the repository, run the compose build, then work at http://127.0.0.1:3000, with data held in the toonflowData named volume mounted at /app/data/ and a myProject working directory initialised on first start.
• Operators running the Web and Server components directly on Ubuntu or Debian with Bun 1.3.14 and ffmpeg, optionally reaching the service from a remote machine over an SSH tunnel.
• Anyone needing additional project directories, which are created from the "select server working directory" dialog or from the command line with docker compose exec toonflow mkdir -p /app/data/workspaces/newProject.

Getting started

Desktop installers for Windows and macOS are published on the GitHub Releases page, while the container route is a clone followed by docker compose up -d --build and a browser at http://127.0.0.1:3000. A server install uses Bun 1.3.14 with bun install --frozen-lockfile, bun run build:server and bun run start:server.

How it compares

The facts name LibTV and TapNow as canvas-based tools that Toonflow approaches in a similar way, and it positions itself as the open-source, self-hosted member of that group. It ships no bundled model capacity: generation is supplied by third-party APIs that the user configures, or by a local ComfyUI and LLM setup, and the README promotes sponsor API platforms with per-use pricing as one way to obtain it.

When to use it — and when not to

Self-hosting carries real operational weight: the Docker path needs Docker Engine and Docker Compose and keeps all state in the toonflowData volume, which means docker compose down -v must be avoided because it deletes that volume, while the server path needs Bun 1.3.14 and ffmpeg installed and running, and neither path supplies model access on its own. macOS users should expect to work through Gatekeeper and quarantine prompts, including the documented sudo xattr -rd com.apple.quarantine /Applications/toonflow.app step. Anyone who wants a fully managed service with bundled credits, or who needs English-first documentation, should look elsewhere: the README body is written in Chinese even though several language headers are offered, and the detailed usage tutorial is hosted as a Feishu document.

sudo xattr -rd com.apple.quarantine /Applications/toonflow.app

sudo spctl --master-disable

git clone https://github.com/HBAI-Ltd/Toonflow-app.git
cd Toonflow-app
docker compose up -d --build

docker compose logs -f toonflow  # 查看日志
docker compose stop             # 停止服务
docker compose start            # 再次启动
docker compose down             # 删除容器，保留数据卷

docker compose exec toonflow mkdir -p /app/data/workspaces/newProject

docker compose stop
docker compose cp toonflow:/app/data ./toonflowBackup
docker compose start

sudo apt-get update
sudo apt-get install -y git curl unzip ffmpeg
curl -fsSL https://bun.com/install | bash -s "bun-v1.3.14"
export PATH="$HOME/.bun/bin:$PATH"
bun --version

git clone https://github.com/HBAI-Ltd/Toonflow-app.git
cd Toonflow-app
bun install --frozen-lockfile

# 构建内置节点、工具、Web 与 Server
bun run build:server
mkdir -p data/workspaces/myProject
bun run start:server