Posted by Kateryna Semenova – Sr. Developer Relations Engineer
AI is reshaping how customers work together with their favourite apps, opening new avenues for builders to create clever experiences. At Google I/O, we showcased how Android is making it simpler than ever so that you can construct sensible, customized and artistic apps. And we’re dedicated to offering you with the instruments wanted to innovate throughout the complete improvement stack on this evolving panorama.
This yr, we centered on making AI accessible throughout the spectrum, from on-device processing to cloud-powered capabilities. Listed here are the highest 3 bulletins you could know for constructing with AI on Android from Google I/O ‘25:
#1 Leverage the effectivity of Gemini Nano for on-device AI experiences
For on-device AI, we introduced a brand new set of ML Equipment GenAI APIs powered by Gemini Nano, our most effective and compact mannequin designed and optimized for operating straight on cell units. These APIs present high-level, straightforward integration for frequent duties together with textual content summarization, proofreading, rewriting content material in numerous types, and producing picture description. Constructing on-device provides vital advantages equivalent to native knowledge processing and offline availability at no further price for inference. To begin integrating these options discover the ML Equipment GenAI documentation, the pattern on GitHub and watch the “Gemini Nano on Android: Constructing with on-device GenAI” discuss.
#2 Seamlessly combine on-device ML/AI with your personal customized fashions
The Google AI Edge platform permits constructing and deploying a variety of pretrained and customized fashions on edge units and helps varied frameworks like TensorFlow, PyTorch, Keras, and Jax, permitting for extra customization in apps. The platform now additionally provides improved help of on-device {hardware} accelerators and a brand new AI Edge Portal service for broad protection of on-device benchmarking and evals. In case you are in search of GenAI language fashions on units the place Gemini Nano just isn’t accessible, you should utilize different open fashions by way of the MediaPipe LLM Inference API.
Serving your personal customized fashions on-device can pose challenges associated to dealing with giant mannequin downloads and updates, impacting the person expertise. To enhance this, we’ve launched Play for On-Machine AI in beta. This service is designed to assist builders handle customized mannequin downloads effectively, guaranteeing the best mannequin measurement and pace are delivered to every Android machine exactly when wanted.
For extra info watch “Small language fashions with Google AI Edge” discuss.
#3 Energy your Android apps with Gemini Flash, Professional and Imagen utilizing Firebase AI Logic
For extra superior generative AI use instances, equivalent to complicated reasoning duties, analyzing giant quantities of knowledge, processing audio or video, or producing pictures, you should utilize bigger fashions from the Gemini Flash and Gemini Professional households, and Imagen operating within the cloud. These fashions are nicely suited to situations requiring superior capabilities or multimodal inputs and outputs. And for the reason that AI inference runs within the cloud any Android machine with an web connection is supported. They’re straightforward to combine into your Android app through the use of Firebase AI Logic, which supplies a simplified, safe approach to entry these capabilities with out managing your personal backend. Its SDK additionally consists of help for conversational AI experiences utilizing the Gemini Stay API or producing customized contextual visible property with Imagen. To be taught extra, take a look at our pattern on GitHub and watch “Improve your Android app with Gemini Professional and Flash, and Imagen” session.
These highly effective AI capabilities will also be dropped at life in immersive Android XR experiences. You’ll find corresponding documentation, samples and the technical session: “The longer term is now, with Compose and AI on Android XR“.
Get impressed and begin constructing with AI on Android at the moment
We launched a brand new open supply app, Androidify, to assist builders construct AI-driven Android experiences utilizing Gemini APIs, ML Equipment, Jetpack Compose, CameraX, Navigation 3, and adaptive design. Customers can create customized Android bot with Gemini and Imagen by way of the Firebase AI Logic SDK. Moreover, it incorporates ML Equipment pose detection to detect an individual within the digital camera viewfinder. The total code pattern is accessible on GitHub for exploration and inspiration. Uncover further AI examples in our Android AI Pattern Catalog.
Choosing the proper Gemini mannequin will depend on understanding your particular wants and the mannequin’s capabilities, together with modality, complexity, context window, offline functionality, price, and machine attain. To discover these concerns additional and see all our bulletins in motion, take a look at the AI on Android at I/O ‘25 playlist on YouTube and take a look at our documentation.
We’re excited to see what you’ll construct with the facility of Gemini!