Google AI Edge Gallery: phone setup and Gemma tutorial

Learn to run Gemma on Android and iOS with Google AI Edge Gallery, adjust thinking settings, import skills, and use image and audio features.

Player not loading? Watch on YouTube

This tutorial walks through Google AI Edge Gallery on Android and iOS, covering model downloads and on-device chat. The presenter describes it as a private way to run models locally and shows how to switch models after initialization.

Hardware advice centers on memory: the speaker recommends 8 GB of RAM or more for Gemma 4 on recent Android phones, an iPhone 15 Pro or newer, or an M-series iPad with 8 to 16 GB. Older phones with 4 to 6 GB are described as better suited to the smaller Gemma 3 1B model. These are the presenter's recommendations, rather than tested compatibility guarantees.

The settings walkthrough covers thinking, temperature, top K and top P. The speaker recommends GPU execution, though the tutorial does not establish that it is faster on every device. A practical limitation is that the demonstrated app retains text input history rather than full conversations.

Agent Skills provide instructions for specific tasks and can be imported from a URL or a local file. The image section shows questions about photos and discusses offline use. Audio Scribe accepts recordings or WAV files; the presenter advises adding a text prompt and resetting the conversation before another clip. Experimental Mobile Actions receives flashlight commands, with a warning that actions may not always work. Prompts Lab supplies presets for summarizing text and rewriting its tone.