Google releases app for testing local AI models

With a new experimental app called AI Edge Gallery, Google enables users to try out various local AI models on smartphones. The application can already be installed on Android devices, with a version for iPhones to follow.

AI Edge Gallery: Try out local AI models on your smartphone

As Google explains on GitHub, the app can help users to find, download and run freely available and compatible local models. After downloading one of the models – such as Google's new Gemma 3n from the AI platform Hugging Face, it is executed locally without the need for an internet connection.

Users can experiment and chat with different models, ask questions about image content, explore prompts, generate and edit code and much more, according to Google.

The Google AI Edge Gallery also offers a so-called “Prompt Lab”, which allows users to perform model-based tasks such as summarizing and rewriting text. The Prompt Lab has several task templates and configurable settings to fine-tune the behavior of the models.

Google points out that the performance of the AI models can vary from device to device. Modern smartphones with more powerful hardware unsurprisingly execute models faster, but model size also plays a role. Larger AI models require more time for a task than smaller models.

Incidentally, the Android app is not available for download from the Play Store, but must be downloaded from GitHub and installed via sideload.

In addition to the experimental app, Google opened up Gemini Nano to third-party developers during I/O. With a new interface called ML Kit, Google's local AI model can summarize and reformulate texts directly on the device.

(afl)