Generative AI is changing how we communicate and create content online. People increasingly turn to AI chatbots like Google Gemini for help with summarizing articles, proofreading emails, or rewriting messages. However, concerns remain about using these tools for sensitive or personal tasks. To address such privacy issues, Google introduced Gemini Nano—a lighter, device-based version of its AI model that runs locally instead of through the cloud.
Until now, access to Gemini Nano was limited to one line of Pixel devices and only supported text input. That’s about to change. Google plans to expand availability and introduce support for image input. This shift aims to make the AI more widely usable while keeping user data private by processing everything directly on the device.
Upcoming Announcements and Developer Tools
In late April, Google revealed the session lineup for I/O 2025, including one titled “Gemini Nano on Android: Generative AI on the Device.” The session description refers to new generative AI APIs powered by Gemini Nano. These APIs will support tasks such as summarizing text, proofreading, rewriting, and generating image captions.
Previously, in October, Google made Gemini Nano accessible via the AI Edge SDK for experimental use on Pixel 9 devices. That SDK enabled text functions like paraphrasing, smart replies, proofreading, and summarization, but didn’t include image captioning. As such, the APIs described for I/O likely point to a new or updated set of tools, distinct from the AI Edge SDK. We won’t have to wait long to learn more.
Just last week, Google quietly released documentation for new ML Kit APIs. ML Kit is a toolkit that simplifies adding machine learning features to Android apps. The new APIs make it easier for developers to use Gemini Nano with out-of-the-box performance via a simplified interface. Like the AI Edge SDK, ML Kit is built on AICore, enabling on-device functionality and improved user privacy.
From Experimental to Beta: Wider Support Ahead
ML Kit’s new GenAI APIs support text summarization, rewriting, and image descriptions—all the features highlighted for the I/O session. Compared to the experimental AI Edge SDK, the ML Kit APIs are offered in “beta,” which likely allows production use in apps.
Currently, developers can’t release apps using the AI Edge SDK, meaning third-party apps haven’t yet leveraged Gemini Nano, notes NIX Solutions. That’s changing. The ML Kit GenAI APIs also go beyond text by supporting image inputs and generating descriptions for images. More importantly, they support a broader range of devices. While the AI Edge SDK was exclusive to Pixel 9, the ML Kit APIs work on any Android phone compatible with the Gemini Nano multi-modal model—such as HONOR Magic 7, Motorola Razr 60 Ultra, OnePlus 13, Samsung Galaxy S25, and Xiaomi 15.
This broader rollout marks a major step in bringing on-device generative AI to more Android users. We’ll keep you updated as more integrations become available.