Google unveiled a covert app this week that lets people run a variety of open AI models from the Hugging Face development platform locally on their smartphones.
Google AI Edge Gallery is the name of the Android application. Soon, an iOS version will be available. The program allows users to look for, download, and run appropriate models to create images, respond to queries, or write and modify code, for instance. On the gadget, everything takes place locally without the need for an internet connection. Supported phones’ processing power is used by the app.
Cloud-based AI models are frequently more potent, but there are drawbacks as well. Some users wish to employ AI in locations without internet access, or they don’t want to transfer sensitive or personal data to outside servers.
App in alpha testing
The AI Edge Gallery is referred to by Google as an experimental alpha version. The program, which is available for download on GitHub, has shortcuts to AI features like conversation and image analysis on the home screen. A list of appropriate models, including Google’s Gemma 3n, appears when you select a function.
Additionally, the software provides a Prompt Lab where users can begin a task with just one input. Rewriting or summarizing text are two examples. The models’ behavior can be altered using the templates and modifiable parameters included in the Prompt Lab. Hugging Face integration allows users to select their own language model, whereas LiteRT allows them to utilize their own language model.
Google cautions that results may differ. The models typically operate faster on newer phones with more potent technology. Another factor is model size: larger models execute tasks more slowly than smaller ones.
Google is asking developers to share their thoughts about AI Edge Gallery. Since the program is licensed under the Apache 2.0 license, there are very few limitations on its use, even in commercial applications.
