Google launches Guided Vision in Gemini Live, turning phone cameras into real-time narrators

Google started pushing Guided Vision to Gemini Live today, giving Android users a way to point their phone's camera at the world and hear Google's AI describe it back in real time. The feature is aimed at small print, unfamiliar surroundings and hard-to-spot objects, and it runs inside the same live camera-sharing mode Gemini already uses for conversational help.

The mechanics are straightforward. You open Gemini Live on a compatible Android handset, share your camera feed, and the assistant narrates what it sees through audio. Google lists reading fine print, describing a room, and finding or identifying objects as the headline uses, which puts the tool closer to a sighted companion than a search box.

That framing matters because Google is not pitching this as a scanner or an OCR utility. Guided Vision keeps the camera live and keeps talking, so the user can move the phone and get updated descriptions rather than a single snapshot result. It is the difference between photographing a menu and having someone read it to you while you decide.

The launch lands in a crowded stretch for camera-based AI. Google has spent the past year pushing Gemini into phones, glasses and other hardware, and rivals including OpenAI and Apple have been building similar visual assistance into their assistants. Guided Vision gives Google a consumer-facing answer to that push, and it ships on Android first, where the company controls the software stack.

Accessibility is the obvious beneficiary. People with low vision have relied on dedicated apps and human volunteers for years to identify objects and read labels, and a built-in option removes a download and a signup from that chain. Whether the descriptions are accurate and fast enough to trust in a grocery aisle or a doctor's office is a separate question from whether the feature exists.

The rollout is limited for now. Google says Guided Vision is arriving on compatible Android devices, without naming a full list, and there is no word on iOS or on when the capability might reach other Gemini surfaces. Users who do not see it today will likely wait on a staged release.

What to watch is how Google handles the hard cases: cluttered scenes, poor lighting, handwriting and privacy. A camera that stays on and streams to a model raises questions about what is processed on-device, what is sent to servers and how long any of it is retained. Google has not detailed those terms yet, and early users will find out fast.