Google DeepMind launches Gemini Sketcher, a vector-first multimodal model for prototyping

AI · 4 min read

DeepMind's Gemini Sketcher is a new model geared toward early-stage prototyping: designers can provide a mix of hand-drawn sketches, voice prompts, and reference screens and receive editable vector wireframes optimized for conversion into Figma or XD files. The model prioritizes structure and semantic layers over pixel-perfect rendering.

Sketcher supports iterative refinement: each pass converts fuzzier sketches into cleaner components while preserving designer intent and offering alternative layouts. The model also annotates suggested interaction flows and provides recommended microcopy for calls to action and error states.

DeepMind emphasized responsible design: Sketcher includes controls to avoid generating plausible but fabricated UI elements (like fake legal disclaimers) and provides provenance metadata to trace how a design evolved from prompts and sketches.