ddcat-ai/open-ai-canvas

影策 (Open AI Canvas)

What it is – An open‑source, web‑based workbench for creating AI‑generated movies, short dramas and other multimedia stories. It lets you move a story from a plain text script to a finished video by stitching together a visual canvas, structured workflow, and a unified task system that calls out to various generative AI models (text, image, video, audio).

Key capabilities

• Free‑form canvas – drag‑and‑drop nodes, connect them, zoom, pan, undo/redo, import/export and read‑only sharing.
• Film‑creation workflow – manage screenplay, characters, scenes, style boards, reference assets, structured story‑boards and a 3‑D director’s view.
• Multimedia generation – asynchronous tasks that produce text, images, video, and audio. Supports reference images, start‑end frames, camera moves, continuation, local edits and batch jobs.
• Task & asset management – queue, progress logs, cancel/retry, media library, reference‑checking and cross‑device sync for logged‑in users.
• Timeline editing – clip arranging, trimming, subtitle transcription and server‑side export, with a plug‑in‑able editing panel.
• Cloud Agent – persistent conversational agent that can summarise the canvas and replay events (currently read‑only).
• Skill ecosystem – built‑in skills for the production pipeline; community‑contributed skill packs can be installed from the judian‑skills repo.
• Admin & usage controls – channel management, usage quotas, feature flags, object storage integration, request interception and an admin console.

Tech stack

• Frontend – React (Zustand for state, localForage for offline cache) served by a Vite‑based dev server.
• Backend – Go 1.25 with Gin, SQLite for local use or PostgreSQL + Redis for production, async workers for model calls.
• Deployment – Docker Compose for local development; official release images (backend, web, agent) for production with optional Host Updater for automatic updates.
• Model integration – “providers” abstract external AI model APIs (e.g., OpenAI, MiniMax H3, Fluxion, etc.) and route generation requests through a secure outbound layer.

Getting started

1. Clone the repo and create a few local cache folders.
2. Run the Go server (backend/cmd/server) with environment variables for address and data directory.
3. In another terminal, install front‑end deps with Bun and start the dev server (bun run dev).
4. Open http://localhost:3000, register an admin account and configure your model API keys in Settings.

Docker users can spin up a full stack with docker compose -f docker-compose.dev.yml up --build (hot‑reload) or build the release images with docker-compose.local.yml.

Safety notes – The README stresses that the default deployment is meant for personal or trusted environments. Public deployments should disable open registration, lock down CORS, keep the backend port internal, and protect API keys and storage directories.

Documentation – The project ships extensive markdown docs covering quick‑start, feature list, code map, local development, database schema, canvas manual, plugin system, and upgrade procedures.

License – MIT.

Who’s behind it – Initiated by ddCat and a community of Chinese‑language contributors; several commercial sponsors provide model‑API services and funding.