Google Launches AI Edge Gallery for On-Device AI Models

Google released the AI Edge Gallery, an experimental alpha application that enables Android users to download and run generative AI models locally on their mobile devices. By utilizing the device's own processor rather than cloud-based data centers, the tool provides offline functionality for tasks such as image generation, coding, and general queries while addressing user concerns regarding data privacy and internet connectivity.

Released under an Apache 2.0 license and hosted on GitHub, the platform integrates with Hugging Face to provide open-source models, including Google's multimodal Gemma 3. The app is built on LiteRT and MediaPipe frameworks and includes specialized tools such as a Prompt Lab for task customization, a RAG library for referencing local documents, and a Function Calling library for automation.

Google noted that performance varies based on hardware capabilities and model size, with larger models requiring more processing time. Because the app requires powerful CPUs and GPUs to manage thermal loads and maintain speed, it may drive consumers toward premium devices. Current limitations include a cumbersome sideloading installation process and varying response accuracy due to the app's experimental status. While currently available for Android, Google plans to release support for iOS in the future.