From Linear Processes to Spatial Organization
Mainstream AI image generation tools, such as Midjourney or Stable Diffusion-based interfaces, typically employ a linear “input-generate-output” model. This approach scatters creative assets like inspiration, reference images, and iterative versions across various file directories or session logs, which is detrimental to overall project management and review.
The Infinite-Canvas project addresses this problem with a spatial solution. It launches a browser-based interface locally at http://127.0.0.1:3000/, centered around an infinitely zoomable and pannable canvas. Users can place images, prompts, workflow nodes, and even entire project assets onto this single canvas, intuitively organizing and managing their creative process through spatial arrangement.
Dual-Mode Design for Diverse User Needs
To accommodate users with different technical backgrounds, Infinite-Canvas provides two core operational modes:
Normal Infinite Canvas
This mode features a node-based editor interface, functionally similar to ComfyUI. Users can construct complex, automated workflows by dragging and connecting different functional nodes—such as image inputs, prompts, LLM processing, loop controls, and API calls. This mode is ideal for professional users who require highly customized and batch-processing tasks.
Smart Canvas
This mode simplifies the complexity of node connections by adopting a more intuitive card-based interaction. Users can directly drag in assets, write prompts, and select a generation method to execute tasks. It introduces a global asset library, allowing users to quickly reference stored reference images using the @ symbol, significantly lowering the barrier to entry for non-technical users.
Highly Open Backend Compatibility
The core design of Infinite-Canvas is to serve as a flexible front-end “command center” rather than a single generation model. It boasts high openness and compatibility, supporting connections to a variety of AI generation backends:
- Cloud APIs: Compatible with nearly all APIs that adhere to the OpenAI standard, including those supporting asynchronous protocols, Google’s Gemini, and other model services like ‘Ark’.
- Free & Commercial Models: Supports connections to free models on the ModelScope platform, as well as CLI tools from Volcano Engine and iDream.
- Local Compute Power: It can directly connect to ComfyUI instances running locally or on a LAN, allowing users to integrate their local GPU power into the canvas workflow. It also supports cloud-based workflow platforms like RunningHub.
Integrated Efficiency Tools and Ecosystem
Beyond its core canvas functionality, the project also integrates a suite of efficiency-enhancing tools to create a closed loop from asset collection to editing and final output.
- Built-in Editing Features: Double-clicking any image on the canvas opens an editing mode that supports cropping, outpainting, mask-based inpainting, video frame extraction, and grid splitting, all without needing to switch to third-party software.
- Workflow Automation: The “Loop Node” supports batch generation across different variables, which is suitable for creating product feature shots, character expression sheets, or storyboards.
- Ecosystem Extensions: The project offers a companion Chrome browser extension for batch-collecting web images and a Photoshop plugin that allows designers to send their work directly to Infinite-Canvas, bridging the gap between design software and the AI generation environment.
The project is open-sourced on GitHub as hero8152/Infinite-Canvas, where users can download a pre-packaged bundle for quick local deployment.
