Google releases Gemini Design API 3.0 with vector-aware generative models

Tech · 6 min read

Gemini Design API 3.0 introduces vector generation capabilities that output editable SVGs and multi-layer assets rather than flattened bitmaps. The model understands constraints like stroke width, path complexity, and color palettes tied to design tokens, which reduces manual cleanup.

Google also added an assets pipeline to the API: clients can specify layout constraints, accessibility rules, and token maps, and Gemini Design will return both raster previews and vector source files. The update supports batch conversion of mood boards into reusable components, with versioning built into the response metadata.

Privacy controls and on-prem options are offered for enterprise customers via Google Cloud AI Workbench. The company emphasized responsible generation practices and provides tools to detect and strip copyrighted embedded elements from training artifacts before deployment.