Google Brings Gemini Nano to Android Apps with New On-Device ML Kit APIs

On May 20, 2025, Google launched a set of on-device GenAI APIs within ML Kit, enabling Android developers to integrate Gemini Nano into their apps. These APIs support summarization, proofreading, rewriting, and image description tasks, all processed locally on the device.

By running entirely on-device, the APIs ensure user data remains private and accessible without an internet connection. This approach also eliminates additional costs per API call. Developers can implement these features without extensive prompt engineering or fine-tuning, thanks to high-level interfaces similar to existing ML Kit APIs.

Also read: Gemini in Gmail Now Handles Google Calendar Tasks on Android and iOS

Each API builds upon the Gemini Nano base model, enhanced with task-specific LoRA adapters and optimized inference parameters for Android. An evaluation pipeline, including LLM raters and human reviewers, ensures quality and accuracy. Benchmark scores indicate significant improvements over the base model, with summarization reaching 92.1 and image description achieving 92.3.

The APIs perform efficiently on devices like the Pixel 9 Pro, processing text at 510 tokens per second and generating output at 11 tokens per second. Image-to-text tasks add only 0.8 seconds for image encoding.

In other headlines, Google also recently held its I/O event with several new announcements. You can read about them here.

• Google I/O 2025: Top 10 Highlights of Google’s New Announcements
• Wear OS 6 Is Launched For Devs, Here’s What’s New in Google’s Smartwatch OS