zhongerxin/cowart

Cowart is a plugin for Codex that gives you an infinite digital canvas right inside your workflow. Think of it as a smart whiteboard where you can sketch ideas, mark up images, and ask an AI to generate or revise images based on your notes and annotations. The tool opens a local web page with a blank canvas space. You can draw, write notes, or upload images to this space. One standout feature is the ability to create a placeholder frame on the canvas, then ask the AI to generate an image that fits perfectly into that frame. You can also take an existing image, draw arrows or write comments on it to show what you want changed, and send that annotated screenshot back to the AI. The AI then reads your notes and produces a clean, revised version of the image, placing it right next to the original without altering your marked-up version. All your canvas work and generated images are saved directly to your own project folder on your computer, not to a cloud service. This tool is built for designers, product managers, or anyone using Codex who wants to think visually. For example, if you are planning a new app, you could lay out rough screens on the canvas, circle a button you do not like, and ask the AI for a better design. Or, if you are writing a story, you could generate character portraits directly into specific spots on your mood board. A key characteristic of this project is that it runs entirely on your local machine. The canvas is powered by an open-source drawing tool called tldraw, and all the data stays in your own project directory. This means your work remains private and portable. The plugin simply bridges the gap between the visual canvas and the AI's image generation capabilities, letting you iterate on designs through a combination of drawing and natural language commands.