Google AI Gallery is a new experimental app that runs AI models on your phone, but can your phone support it?

In a surprising move, Google has launched an experimental application named AI Edge Gallery, enabling Android users to execute various open-source artificial intelligence models directly on their devices without needing an internet connection. Currently available as an alpha version on GitHub, AI Edge Gallery permits users to download and utilize compatible AI models for diverse tasks such as image analysis, chat interactions, text generation, and code editing, all while operating offline. This innovation allows the models to leverage the phone’s CPU, GPU, or other processing units, enhancing accessibility in offline scenarios and addressing privacy issues often linked with cloud-based solutions. Although the app is not yet accessible through the Google Play Store and requires sideloading, it offers access to open-source models from platforms like Hugging Face, including Google's newly introduced Gemma 3n. Once deployed, these models can perform tasks like image analysis, answering questions, rewriting text, and generating code, functioning seamlessly even in airplane mode. AI Edge Gallery incorporates three primary tools, with the performance of these models being significantly influenced by the device’s hardware capabilities. More demanding models and intricate tasks necessitate more powerful processors, which may pose challenges for older or mid-range smartphones regarding speed and thermal management. Google provides users with the option to select the processor (CPU, GPU, etc.) for the model execution, track temperature, and modify configuration settings such as tokens and temperature values. Users should be mindful of potential battery drain and overheating issues during prolonged usage, especially on devices with less robust hardware. This initiative highlights Google’s ongoing commitment to bringing AI closer to the user’s edge, focusing on personal devices rather than solely depending on cloud computing. Although the app's on-device models may not yet match the prowess of comprehensive tools like ChatGPT or Gemini, it represents a noteworthy advancement towards achieving more private, efficient, and accessible AI interactions. Presently, the AI Edge Gallery app is available under the open-source Apache 2.0 license, enabling both academic and commercial applications. While it is currently exclusive to Android users, Google has announced that an iOS version is in the pipeline. Developers and early adopters eager to explore local AI functionalities can download the APK and follow installation guidelines on Google’s GitHub page, although those less experienced with sideloading may prefer to wait for the app's official launch on the Play Store for enhanced security and ease of use.