Experimental hybrid inference and new Gemini models for Android
Android developers can now implement advanced AI features thanks to new updates to Firebase AI. Hybrid inference allows dynamic switching between on-device and cloud-based Gemini models via a unified API. This API supports Gemini Nano on-device and cloud Gemini models from Vertex AI and the Developer API. Developers can use PREFER_ON_DEVICE or PREFER_IN_CLOUD modes during GenerativeModel initialization. The Firebase hybrid inference API is experimental, available through specific dependencies, and ideal for Firebase AI Logic users. A new sample demonstrates review generation and translation using the hybrid inference. New Nano Banana models, including Pro and 2, are available for image generation with improved capabilities. The updated Magic Selfie sample now uses Nano Banana 2 for easier background changes. Gemini 3.1 Flash-Lite, a new version of the Flash-Lite family, is also available for those with usage examples like in-app translation. The Gemini models are easy to integrate via the Firebase AI Logic SDK. Developers are encouraged to explore the Hybrid sample and documentation for these new features.
PREFER_ON_DEVICEorPREFER_IN_CLOUDmodes during GenerativeModel initialization. The Firebase hybrid inference API is experimental, available through specific dependencies, and ideal for Firebase AI Logic users. A new sample demonstrates review generation and translation using the hybrid inference. New Nano Banana models, including Pro and 2, are available for image generation with improved capabilities. The updated Magic Selfie sample now uses Nano Banana 2 for easier background changes. Gemini 3.1 Flash-Lite, a new version of the Flash-Lite family, is also available for those with usage examples like in-app translation. The Gemini models are easy to integrate via the Firebase AI Logic SDK. Developers are encouraged to explore the Hybrid sample and documentation for these new features.