Scribble Diffusion is an open-source web app from Replicate that turns a rough hand-drawn sketch into a polished, detailed image using AI. You draw on a canvas in the browser, add a text prompt describing what the sketch represents, and a ControlNet-based image model refines your doodle into a finished picture while following its shapes. The project has close to 3,000 stars on GitHub, and a live demo runs at scribblediffusion.com.
The app is a small, readable Next.js project that pairs a simple drawing canvas with Replicate’s hosted implementation of the scribble ControlNet model. Because the sketch conditions the generation, the output respects your composition — a childlike house sketch becomes a realistic house in the same position and proportions. It is one of the clearest minimal examples of how to build an interactive AI image product: the entire application, from the canvas UI to the API integration, fits in a compact codebase that developers routinely fork as a starting point for their own image tools. Swapping in a different ControlNet model or changing the interface takes only a few lines of code.
Scribble Diffusion is free and open source under the MIT license. Self-hosting uses Replicate’s API for the model, which is billed per generation at Replicate’s standard rates; the public demo is free to try.
Scribble Diffusion is for curious creators who want to play with sketch-to-image generation, and for developers looking for a clean reference implementation of a ControlNet-powered web app.