Stability AI Launches 'StableUI' for Editable, Layered Design Generation

AI · 6 min read

Stability AI's StableUI is trained to produce layered design outputs rather than flat images, returning structured files with named layers, masks and suggested tokens. The model supports prompts for styles, platforms, and accessibility targets and focuses on generating outputs suitable for iteration.

Unlike previous generative models that output raster images, StableUI attempts to reason about hierarchy and export results as vector-friendly, editable packages which can be imported into popular design tools. The company claims a lower post-edit overhead and fewer manual cleanups.

Stability AI is releasing developer APIs and a community license that encourages integration into open-source prototyping apps. They also published a whitepaper detailing how dataset curation and masking techniques reduce hallucinated UI elements.