Google brings live video and screen-sharing to Gemini Live on Android

What Google actually demoed on Android Avenue

Google used its MWC booth, positioned on "Android Avenue" between Halls 2 and 3, to show two existing features and one new one. The existing pieces were Gemini Live handling complex topics across multiple languages and Circle to Search translating a restaurant menu.

The genuinely new item is live video and screen-sharing inside Gemini Live. That is a change in input modality: instead of asking about a single captured image, a user can point a running camera at the world or share what is on their screen, and Gemini responds against that continuous stream.

The two demos reveal the intended use case

Google chose consumer-styling scenarios to illustrate the feature. In the video demo, a person films various pottery glazes and asks which suits a mid-century modern look; based on the finished vases it sees, Gemini suggests olive green or muted blue.

In the screen-sharing demo, a user shares a photo of jeans from their screen and asks for outfit ideas. Gemini proposes a white T-shirt or turtleneck, then, asked about a jacket, recommends denim or bomber.

Both examples are low-stakes aesthetic advice. That is a deliberate framing choice. The demos show the model reasoning over ambiguous visual context rather than fetching a definite answer, which suits subjective questions where a wrong suggestion carries little cost.

The subscription gate is the operative detail

...new live video and screen-sharing capabilities in Gemini Live, which will start rolling out to Gemini Advanced subscribers as part of the Google One AI Premium plan on Android devices later this month.Montana Labs

The feature is not a broad Android rollout. It is scoped to Gemini Advanced subscribers on the Google One AI Premium plan, on Android, starting later in March. Continuous video and screen inference is expensive to serve, so restricting it to paying subscribers aligns the compute cost with a revenue stream.

The partner-device angle matters here too: Google notes the latest Android partner devices bring these experiences to life, tying the software capability to hardware Google does not itself sell.

What continuous visual input changes for anyone building on Android

The shift from single-image to live-stream input is the part teams should track. A snapshot is one API call against one frame; a video or shared-screen session implies sustained inference, session state, and latency budgets that behave differently under real network conditions on a show floor or a street.

Google's decision to launch this behind the Advanced subscription rather than the free tier is a signal about the cost profile of always-on multimodal assistance. For anyone planning similar features, the constraint to plan around is not model capability but the per-session economics of keeping a camera or screen feed running against a large model.