Generate an image inside Codex and place it directly onto a persistent local canvas.
Draw arrows and notes on a generated image to tell the AI exactly what to change.
Iterate on the same image across several rounds while keeping earlier versions visible side by side.
Keep one reusable canvas of generated images and annotations saved locally on your machine.
| tasihi89/canvas-hand | siigari/claude-heartbeat | 7-e1even/learn-agent | |
|---|---|---|---|
| Stars | 165 | 161 | 160 |
| Language | JavaScript | JavaScript | JavaScript |
| Setup difficulty | easy | moderate | easy |
| Complexity | 2/5 | 3/5 | 2/5 |
| Audience | vibe coder | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires Codex with plugin support, dependencies for the local canvas service install automatically on first use.
Canvas Hand is a plugin for Codex that adds a local infinite drawing canvas, built on the tldraw library, to Codex's own sidebar browser. It lets you generate images with Codex's built-in image model, place them on the canvas, and then draw arrows and notes directly on an image to tell the AI what to change. Codex reads those annotations and redraws the image accordingly, placing the new version next to the original rather than replacing it, so you can keep iterating on an image over several rounds. The canvas runs as a small local web service on your own machine, by default at address 127.0.0.1 port 8000, and its data is saved to a fixed folder on your computer, so you always come back to the same canvas. Only the arrows drawn on the most recently generated image are used when you ask for a redraw, which keeps each round focused on the image you just made. Installing it means cloning the plugin's code to a folder on your machine and registering it with Codex, either by asking Codex to do the setup itself, since the plugin includes instructions written for that purpose, or by manually editing two configuration files, a marketplace file and a config file, then restarting Codex. No separate install step is needed for the canvas's own dependencies, since they install automatically the first time you open it. Once installed, you open the canvas by asking Codex in plain language, generate an image by describing what you want, and revise an image by drawing an arrow with the canvas's "revise instruction" pen and telling Codex to redraw based on your annotations. The project ships as a set of three focused skill files that Codex routes between depending on what you are asking for.
Canvas Hand adds a local drawing canvas to Codex, letting you generate images, mark them up with arrows, and have the AI redraw them.
Mainly JavaScript. The stack also includes JavaScript, tldraw, Vite.
No license information is stated in the README.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly vibe coder.
This repo across BitVibe Labs
Verify against the repo before relying on details.