Google launches Gemini Canvas: a multimodal model for UI generation

AI · 5 min read

Google announced Gemini Canvas, a multimodal model tailored to UI generation that accepts text prompts, screenshots, and hand-drawn sketches. Its architecture prioritizes semantic understanding of interface elements to produce editable layouts and accessibility suggestions.

Gemini Canvas includes a new 'contrast-aware' mode that flags low-contrast text, suggests accessible color swaps, and generates accessible style tokens. Google says the model was trained with consults from accessibility experts to reduce common pitfalls in automated UI generation.

The company is offering plugins for major design environments and APIs for automated layout conversion. Gemini Canvas aims to be a backend model for designers while Google continues to stress shared ownership of final designs and human-in-the-loop validation.