#GoogleIO 2025 Recap

Key Concepts

Project Astra, Gemini API, 2.5 Flash native audio model, Compose Adaptive Layouts, Android XR, Project Moohan, Android Studio AI agent, Gemini, Carousels, Hover cards, Gemini Nano, Multimodal AI APIs, Firebase Studio, Builder.io, Gemma 3n, SignGemma, American Sign Language (ASL).

Gemini API and Audio Model Updates

Logan Kilpatrick introduces updates to the Gemini API, highlighting the new 2.5 Flash native audio model. This model is improved in its ability to ignore extraneous sounds and voices, enhancing audio clarity. It also supports 24 languages natively, expanding its global usability. Paige Bailey demonstrates the API's capabilities by asking Gemini to locate Shoreline Amphitheatre on a map and then find nearby coffee houses with Wi-Fi.

Adaptive Layouts and Android XR

Diana Wong discusses advancements in adaptive layouts using the Compose Adaptive Layouts library, specifically mentioning pane expansion. She emphasizes the importance of building adaptively for Android XR, the extended reality platform developed in collaboration with Samsung. Project Moohan, a Samsung headset powered by Android XR, is mentioned as a platform developers can start building for immediately, with consumer availability expected later in the year.

AI Agent in Android Studio

Florina Muntenescu demos a new AI agent integrated into Android Studio. This agent is designed to assist developers with version updates. The agent attempts to build a project and then uses Gemini to identify and resolve build errors. The process is iterative, with the agent continuously working to achieve a successful build.

UI Element Enhancements

Una Kravets showcases new capabilities that simplify the creation of complex UI elements such as carousels and hover cards.

Gemini Nano and Multimodal Capabilities

Addy Osmani introduces new multimodal capabilities powered by Gemini Nano. These built-in AI APIs enable users to interact with Gemini using both audio and image input. He demonstrates this by using a webcam to take a photo of a theater ticket. The AI instantly identifies the seat section based on the image.

Figma to Firebase with Builder.io

David East explains how Figma designs can be brought to life in Firebase Studio using Builder.io. The Figma export generates well-isolated components instead of a monolithic code block. Updating the Figma file and refreshing the webpage results in a fully functional product detail page.

Gemma 3n and SignGemma

Gus Martins announces Gemma 3n, a model that can run on as little as two gigabytes of RAM. This makes it faster and more efficient on mobile hardware compared to Gemma 3. Gemma 3n also includes audio understanding, making it multimodal. SignGemma, a new family of models, is introduced for translating sign language to spoken language text, with a focus on American Sign Language (ASL) to English translation.

Synthesis/Conclusion

The presentation highlights several advancements in AI and Android development, including improved audio processing with the Gemini API, enhanced adaptive layouts for Android XR, AI-assisted version updates in Android Studio, multimodal capabilities with Gemini Nano, streamlined design-to-deployment workflows with Figma and Firebase, and efficient on-device AI with Gemma 3n. The introduction of SignGemma also demonstrates a commitment to accessibility through AI-powered sign language translation. These updates collectively aim to empower developers to create more intuitive, efficient, and accessible applications across various platforms.