Google has rolled out a set of major updates to its Gemini app, aiming to widen its appeal and offer users more ways to interact with AI. Announced at the annual Google I/O 2025 event on Tuesday, the updates include tools for visual help, media generation, and research — now available on both Android and iOS. Here is every AI update announced by the American tech giant:
Gemini Live
Gemini Live, which lets users share their camera feed or screen during conversations, is now available for free. The feature is designed to help users show what they mean instead of typing out questions. According to Google, conversations using Gemini Live tend to be longer than text-only chats, which the company attributes to the more interactive format.
Google plans to integrate the tool more deeply into its wider ecosystem in the coming weeks. Users will be able to link Gemini Live with apps like Maps, Calendar, Tasks, and Keep. For example, asking about restaurant options may link directly to Google Maps, or a group chat could lead to an event being added to Calendar.
Imagen 4
Gemini now includes Imagen 4, a new image generation model that supports improved visual detail and better text rendering within images. Users can create graphics and visuals for various uses, including presentations and social media posts.
Veo 3
For video creation, Veo 3 is being introduced. It supports text-to-video generation and can also add ambient sounds and basic character dialogue. Veo 3 is currently available only toGoogle AI Ultra subscribers in the U.S., limiting access for international users or those on the free plan.
Deep Research
The Deep Research feature now allows users to upload personal files — such as PDFs or images — to be included in AI-generated reports. The goal is to provide more personalised and context-rich results by combining private and public data sources. Google has announced plans to expand this functionality to include content from Google Drive and Gmail in the near future.
Project Astra
Project Astra showcases the real-time capabilities of Google’s Gemini models, with the initial features now integrated into Gemini Live. This upgraded version can actively use a device’s camera to interpret on-screen content in real-time. Among the latest improvements are a more natural, expressive voice powered by native audio generation, enhanced memory functionality, and advanced computer control features.
During the keynote at Google I/O 2025, a live demonstration highlighted Gemini Live’s ability to interact fluidly with users—responding with expressive speech, handling interruptions seamlessly, and continuing conversations without losing context. It also showcased multitasking abilities such as making business calls, scrolling through documents, and browsing the web, all in real-time.
Google Flow
Google Flow is an AI-powered filmmaking tool designed for creatives to effortlessly generate cinematic videos. It combines Google’s advanced models — Veo (video generation), Imagen (image generation), and Gemini (natural language understanding) — to help users turn everyday language prompts into high-quality visual scenes. Flow enables consistent character and scene creation, allowing seamless integration across multiple clips. It’s built to make storytelling faster, intuitive, and visually stunning.
Agent Mode
At the Google I/O 2025 event, Pichai unveiled a new feature called Agent Mode for the Gemini app. This upcoming experimental tool, initially available to subscribers, is designed to handle complex tasks and planning on the user’s behalf. With Agent Mode, Gemini moves beyond simple responses to take on more autonomous actions—organising, scheduling, and executing multi-step tasks. Google also announced that these agentic AI capabilities will extend to Chrome, Search, and the Gemini platform, marking a significant step toward AI that can proactively manage tasks rather than just react to prompts.
Google Jules
Jules is an autonomous, agentic coding assistant that works directly with your codebase. Unlike traditional code-completion tools, Jules clones your repository into a secure Google Cloud VM, understands your project context, and independently handles tasks like writing tests, fixing bugs, building features, and more. It works asynchronously, so you can focus elsewhere while it completes tasks and returns with a detailed plan, reasoning, and code changes. Jules is now in public beta and prioritises privacy, keeping your code secure and isolated.
AI Mode in Search
Google is rolling out a new feature called AI Mode, aimed at people who want a more advanced and interactive search experience. First tested in Labs, AI Mode is now available to everyone in the U.S., with a wider global rollout expected later. A new tab for AI Mode will soon appear in the Google app and on desktop.
AI Mode uses something called a “query fan-out” system. This means it breaks down your question into smaller parts and runs many searches at once, helping it dig deeper and return more useful and detailed answers from across the internet. It also uses Gemini 2.5, Google’s most advanced AI model yet.
With AI Mode, users can ask follow-up questions, get interactive links, and even use images or live video to search in real-time. It is not just about answering questions anymore, Google wants to help peopledo things, from booking tickets to comparing data.
Google Meet speech translation in real-time using AI
Google has introduced a groundbreaking AI-powered speech translation feature in Google Meet, enabling real-time audio-to-audio translation during calls. Built on DeepMind’s advanced AudioLM technology and integrated with the Gemini AI model, this system translates spoken language into a listener’s preferred language—while preserving the speaker’s original voice, tone, and emotional expression.
Unlike traditional caption-based translation, this feature directly transforms speech, delivering natural-sounding audio in real time. Users hear the translated voice with subtle overlays of the original, enhancing clarity and maintaining conversational context. Though there is a slight delay for processing, the experience closely mimics having a live interpreter on the call.
Google Beam
Google Beam is a new 3D video communication platform that transforms regular 2D video calls into immersive 3D experiences. Announced at Google I/O 2025, it uses multiple cameras and AI to create realistic, real-time 3D visuals with depth and eye contact. Powered by Google Cloud and designed for enterprise use, Beam also supports precise head tracking and is expected to feature real-time speech translation. It will roll out on HP devices later this year.
Gemma 3n
Gemma 3n is Google’s first open AI model designed specifically for on-device use, bringing fast, multimodal intelligence to phones, tablets, and laptops. Built on a new architecture developed with partners like Qualcomm, MediaTek, and Samsung, it powers real-time, private AI experiences without relying on the cloud. Gemma 3n also forms the foundation for the next generation of Gemini Nano, enabling developers to explore advanced AI directly on everyday devices.
Try-on
The “Try On” feature in Google Search allows users to see how clothes like shirts, dresses, pants, and skirts would look on them by uploading a full-length photo. Available through Search Labs in the U.S., the tool uses AI to generate a visual of the outfit on the user. It also lets users save or share the images for feedback before making a purchase.
