Google launches Gemini Micro: tiny transformer for mobile prototyping

Tech · 3 min read

Gemini Micro targets designers and developers who need fast, private inference for tasks that don’t require full cloud models. Google optimized the model for on-device latency and low memory, and released runtimes for Android and iOS as well as a Figma plugin that runs inference locally where feasible. The product brief emphasizes tightly constrained context windows and ergonomics for iterative design work.

The Figma integration surfaces inline copy suggestions, alt text generation, and conditional microcopy tied to component states. Android Studio’s integration focuses on intent parsing and localized string suggestions, aiming to reduce time spent writing UX edge-case prompts. Google also published a best-practices guide for hybrid workflows where Gemini Micro seeds content and a cloud model refines it.

Privacy-conscious teams appreciated the option to run sensitive flows entirely offline. Performance tests show Gemini Micro outperforms cloud round-trips for frequently repeated lightweight tasks, while Google plans to maintain a seamless fall back to larger Gemini models for heavier multimodal work.