Google has launched an experimental application called Google AI Edge Gallery, now available on the Google Play Store. The app integrates Google’s self-developed lightweight multimodal models, Gemma3 and Gemma3n, enabling a range of on-device AI capabilities including image recognition, audio transcription and translation, and text-based conversations. All processing occurs locally on the user’s device, ensuring full offline operation and enhancing data privacy. Gemma3n, a key component of the app, is built using the Matryoshka Transformer architecture, which allows for efficient handling of long context lengths and multilingual tasks. This design enables high-performance AI inference on mobile devices with minimal power consumption, making advanced AI features accessible even on resource-constrained hardware. The launch marks a significant step in Google’s push toward decentralized, privacy-preserving AI, demonstrating how powerful multimodal models can run effectively on consumer devices without relying on cloud connectivity. By bringing AI capabilities directly to the edge, Google aims to deliver faster, more responsive experiences while keeping user data secure and private. The application is part of Google’s broader effort to expand the reach of its AI models beyond data centers, empowering developers and users with tools that prioritize performance, efficiency, and privacy.