Most visual-iteration workflows force creators to switch between separate generation UIs, asset managers, and note systems. By embedding model-driven generation, reference editing, and node-based canvas composition into a single browser front end, this project lets you prototype visual pipelines and iterate images continuously without context-switching.
What Sets It Apart
- Integrated generation + composition: connect image/video generation nodes directly to canvas nodes so model outputs become first-class elements you can rearrange, chain, and refine — this reduces friction when exploring variants.
- Frontend-direct API flexibility: the app connects to user-configured OpenAI-compatible endpoints (supports adapters like chatgpt2api, grok2api, flow2api, newapi), so teams can swap providers or local proxies without changing the UI stack.
- Local agent and plugin extensibility: ships with a local Canvas Agent and a plugin/TypeScript SDK (MCP integration) for automating node operations or adding custom node types, enabling programmatic workflows and multi-agent collaboration.
- Lightweight persistence model: configuration, keys, and generation history are stored in-browser by default, which simplifies local testing and privacy but limits out-of-the-box multi-user production readiness.
Who It's For and Trade-offs
Great fit if you are a visual creator or designer who iterates on image ideas, a solo developer wanting an extensible front-end to stitch model calls into pipelines, or a team prototyping AI-assisted creative flows without committing backend infrastructure. Look elsewhere if you require hardened multi-tenant hosting, enterprise-grade access control, or stable backward-compatible storage today — the project is actively developed, uses browser-local storage by default, and is licensed under AGPL-3.0 which has copyleft implications for hosted derivatives.
