Infinite Canvas

Overview

Infinite Canvas is an open-source workbench for image and video creation. It brings canvas orchestration, AI image generation, reference-image editing, a chat assistant, a prompt library, and asset management into one interface. Users can generate, connect, and recompose images, text, and shapes on the canvas, turning creation from single generations into continuous iteration. It supports visual creation workflows and multi-Agent collaboration.

Key Features

• Infinite canvas: multiple canvas projects, node drag and zoom, connections, minimap, undo and redo, and import and export.
• AI creation: the browser front end connects directly to configured OpenAI-compatible interfaces for text-to-image, image-to-image, reference-image editing, text Q&A, audio, and video generation; compatible with channels such as chatgpt2api, grok2api, flow2api, and newapi.
• Canvas assistant: converse and generate images around selected and upstream nodes, then insert results back into the canvas.
• Local Agent and Codex app plugin: connect Codex or Claude Code through the local Canvas Agent so the Agent can operate the current canvas via MCP; the Codex plugin registers MCP automatically and attempts to launch the local Agent.
• Plugin system and custom interfaces: install, enable, update, and uninstall remote node plugins by URL; use the TypeScript SDK to develop canvas node plugins; customize image and video interface calls.
• Prompt library: includes seven open-source prompt sources and supports custom standard JSON sources, connected directly by the browser front end and cached in IndexedDB.

Use Cases

• Exploring visual directions and iterating continuously: generate, connect, and recompose images, text, and shapes on the same canvas, and save reusable prompts, reference styles, and result images.
• Creating posters, presentations, infographics, and other visual assets: prompt library examples cover Apple-style posters, Zootopia posters, gradient glass-style PPT slides, cartoon infographics, and city posters.
• Reference-image editing and style replication: edit with reference images, including generating posters from reference styles, replicating PPT styles, and applying the same filter effect to photos.
• Video generation and local Agent collaboration: generate video, and connect Codex or Claude Code through the local Canvas Agent to operate the current canvas via MCP.

Getting Started

• For local development, clone the repository, enter the web directory, run bun install, and then run bun run dev.
• For Docker, run docker compose up -d in the repository root; the default port is 3000.
• After first opening the app, go to settings in the top-right corner and enter your OpenAI-compatible Base URL and API Key. If the default OpenAI interface call method does not match your API, customize the image or video script calls.

Deployment & Requirements

• A working OpenAI-compatible interface is required, with Base URL and API Key configured.
• AI API keys, Base URLs, canvas data, assets, and generation records are stored in the browser locally by default.
• Docker deployment uses default port 3000.
• The browser front end connects directly to configured OpenAI-compatible interfaces.

Before You Adopt

• License: MIT. Review its terms before using, modifying, or distributing the project.