saito-sakuya/candycraft

Candy Craft is a visual tool for building and refining prompts used with AI image generation services. Rather than writing a prompt entirely by hand, you use a canvas-based interface to arrange scene elements, adjust composition, configure lighting, and then export a polished prompt you can paste into an image generator. The canvas works like a simple diagram editor: you drag elements around, set which ones appear in the foreground versus background, establish links between them, and specify any text that must appear exactly as written in the final image. The tool tracks eight adjustable style dimensions and shows them as a radar chart so you can see the balance visually. A composition panel lets you set the aspect ratio and final pixel size. A lighting and camera panel lets you position light sources from three views (top, front, side) and see a 3D preview of the scene layout. Once the canvas is set up, you click a button to analyze the current state and generate an optimized prompt. The tool sends the canvas data to an AI model, which writes the final prompt in a structured format and returns it for you to copy. All the AI output goes through a versioned schema internally, which means the tool validates the response before showing it to you and falls back gracefully if something goes wrong. The project comes in two operating modes. A backend-hosted mode routes AI calls through server-side functions on Cloudflare Pages, keeping your API keys private. A user-configured mode lets you point the tool at any OpenAI-compatible API endpoint using keys you store in your own browser. The project is entirely static HTML and JavaScript with no build step required. Deployment to Cloudflare Pages requires pushing the repository to GitHub and connecting it in the Cloudflare dashboard. No build command is needed. The tool comes with two visual themes (Candy and Pro) that can each be used in light or dark mode. The project is licensed under the MIT license.