Stability AI releases DreamLayout: a diffusion model for layout generation

AI · 3 min read

Stability AI's DreamLayout repurposes diffusion techniques for structural design: instead of pixels, the model generates layout primitives like columns, gutters, and card placements. Designers provide context — wireframes, content counts, or style hints — and DreamLayout proposes multiple layout samples ranked by visual balance and information density.

The model includes a sampling control panel that lets designers prioritize responsiveness, compactness, or whitespace, and it can output JSON layout specs that map directly to design tooling and frontend layout systems. DreamLayout is available on Hugging Face and via an open-source inference toolkit for local use.

Stability emphasized community involvement: users can contribute layout datasets and reward patterns back to the model's training pool. The release also includes a small GUI demo for rapid concept testing and an export plugin for Figma and Sketch.