Google announced new updates to its AI products at the Google I/O 2025 event. These include improvements to the Gemini app, new models for video and image creation, and fresh tools for creators and developers.
Gemini Live: Camera, Screen Sharing, and Ecosystem Integration
Beginning May 20, 2025, Gemini Live allows users to use their phone camera to interact with objects and talk about them in real time, with free camera and screen sharing available on both Android and iOS.
Soon, Gemini Live will integrate with daily apps — it can add events to Google Calendar or fetch local info like pizza options from Google Maps. Users can control app connections and data via settings.
Veo 3, Imagen 4, Lyria 2, and Flow: AI for Creators
-
Veo 3: Google’s newest video AI model produces high-quality videos with synchronized audio, including background sounds and character dialogue. It understands prompts well, turning text stories into video clips. Available now for Ultra subscribers in the U.S. and enterprise users on Vertex AI.
-
Veo 2 updates: Added features include reference-powered video (use images for style control), camera movements, outpainting (wider frames), and object add/remove functions. These are accessible now in Flow and will arrive in Vertex AI soon.
-
Flow: A filmmaking AI tool that combines Veo, Imagen, and Gemini models to create cinematic scenes from natural language descriptions. U.S. subscribers to Google AI Pro and Ultra have access to this feature.
-
Imagen 4: Google’s latest image generation model delivers sharp details at up to 2K resolution in various styles. It also improves text accuracy for creating cards, posters, and comics. Available today in Gemini, Whisk, Vertex AI, and Google Workspace apps. A faster version will launch soon.
-
Lyria 2: Expanded access to AI music tools in YouTube Shorts and Vertex AI. Lyria RealTime allows real-time interactive music creation, available via API and AI Studio.
Deep Research and Gemini App Enhancements
- Deep Research: Starting May 20, users can upload private documents like PDFs and images to generate customized research reports that combine personal and public data. Upcoming support includes Google Drive and Gmail integration.
- Canvas in Gemini: An interactive creative space enhanced with Gemini 2.5 models, enabling users to build infographics, quizzes, and podcast-style audio in 45 languages. It also converts descriptions into working code rapidly.
- Gemini in Chrome: U.S. Google AI Pro and Ultra subscribers will get Gemini on desktop starting May 21, offering web content summaries and clarifications, with multitasking across tabs coming later.
Interactive Quizzes and Gemini 2.5 Model Updates
- Quizzes: Gemini now supports creating practice quizzes with instant feedback, adapting follow-up questions to improve learning. This feature is globally available.
- Gemini 2.5 Pro: Updated for enhanced web app building, coding, and long-context understanding with a 1 million-token window. It leads academic and coding benchmarks and is preferred by educators.
- Deep Think mode: An experimental reasoning feature that evaluates multiple hypotheses before answering. It scores high on difficult math and coding tests and will be released to trusted testers soon.
- 2.5 Flash: A faster, more efficient model available for preview and general release in June, showing improvements in reasoning, code, and multimodal tasks.
New Gemini 2.5 Capabilities
- Live API updates: Preview of audio-visual input and native audio output, supporting tone, accent control, and tool use. Features include emotion detection (Affective Dialogue), background noise filtering (Proactive Audio), and enhanced reasoning (Thinking in Live API).
- Text-to-speech: Now supports multiple speakers with expressive voices, available in 24+ languages.
- Computer use: Integration of Project Mariner’s task management abilities into Gemini API and Vertex AI, with several partners testing it.
- Security: Improved defenses against indirect prompt injections, making Gemini 2.5 the most secure model family so far.
- Developer tools: Thought summaries organize the model’s internal reasoning for better clarity. Thinking budgets let developers control token usage and latency. Native SDK support for Model Context Protocol (MCP) improves integration with open-source tools.
Google AI Pro and AI Ultra Subscription Plans
-
Google AI Pro ($19.99/month): Includes the Gemini app with added features, Flow filmmaking tools, NotebookLM, and early access to Gemini in Chrome.
Free access is extended to university students in the U.S., Japan, Brazil, Indonesia, and the United Kingdom.
-
Google AI Ultra ($249.99/month): Offers highest usage limits and access to top-tier models and features, including Veo 3 video, Deep Think mode, advanced Flow controls, Whisk Animate, NotebookLM, and integration across Google apps. It also includes YouTube Premium and 30 TB of storage.
Now offered in the U.S. with a 50% introductory discount for the first three months.
Responsible AI and Content Verification
Google has applied SynthID watermarks to over 10 billion AI-generated images, videos, audio, and text since 2023 to help track AI content and reduce misinformation. Now, SynthID Detector is available to verify if a file contains this watermark, helping users confirm AI origin.
You can give Gemini a try at gemini.google.com.