Gemini Nano: Google opens up local AI for app developers

Google is bringing local AI to more Android devices. With the new ML Kit GenAI interface, third-party apps can also access the on-device AI Gemini Nano to use writing aids and more. However, smartphones with weaker computing power are apparently excluded for now.

The new Gemini Nano interface ML Kit GenAI for third-party providers has not yet been officially announced by Google, but the company has already published the developer documentation. It was first discovered by Android Authority and is expected to be officially announced at Google I/O 2025 on May 20.

Gemini Nano: Writing aids, summary and more directly on the Android device

According to the Documentation to ML Kit GenAI, developers will be able to access the local AI Gemini Nano for their apps to help users with basic tasks. The new AI functions are based on AICore, an Android system service that enables the execution of basic generative AI models on the device.

The functions offered by Gemini Nano are roughly similar to those that Apple has integrated with iOS 18.4 and macOS in its system apps such as Mail, Notes, and others with Apple Intelligence. They include text summaries, correction suggestions or image descriptions and the option to rephrase short messages in different tones or styles. Samsung also offers similar features on some smartphones with Galaxy AI.

According to Google, not all the new Gemini Nano functions will be available for all languages at first. For example, the summary function, which can create up to three key points from a text or chat history, is only available in English, Japanese, and Korean. The proofreading and rewriting functions support seven languages – including German. The image description function is initially only available in English.

The advantage of local AI is obvious: personal content is not sent to Google's cloud server, which is a benefit for privacy. Another positive effect of local AI is likely to be a higher output speed, as data does not have to be sent back and forth. The new AI functions can also be used offline – for example in flight mode or in a dead zone –. According to Google, there are no additional server costs for developers for each API call.

Gemini Nano: ML Kit GenAI apparently requires high on-device computing power

The new on-device AI functions require sufficient computing power and memory on the device, which apparently neither entry-level nor mid-range smartphones fulfill. In addition to its current top models from the Pixel 9 series, Google also mentions the high-end models from Samsung's Galaxy S25 series and Xiaomi's 15 series models as well as top devices from Oppo, Oneplus, Honor and Motorola.

Even Google's new mid-range Pixel 9a model is not included in the list of supported smartphones. At least Google mentions that support will be extended to other devices.

The new interface appears to replace or supplement Google's experimental AI Edge SDK previously used for the Pixel 9 series. This was previously only used for the Pixel 9 series.

(afl)