Gemini Live Video Hands-On: The World-Facing Camera We’ve Waited For [Video]

Google has been rolling out Gemini Live’s visual Astra-powered video functions super slowly, but it is starting to expand. Here’s what it’s like to use and what you need to know.

The company is suggesting that the arrival of the camera and screen sharing controls are part of an “April 2025 Pixel Drop” but a non-scheduled set of features feels odd given that the function will work on practically any Android phone.

When you launch the Gemini app on your Android phone, a pop-up will indicate that Gemini Live’s Astra-powered features are ready to test. The mini pop-up says you can “talk through ideas, learn about your environment” or “get help with what’s on your screen.” The latter refers to the screen-sharing function, which is like a souped-up version of Circle to Search.

Gemini Live Astra: How to access

To access any of the new visual modes, you will need a Google One AI Advanced subscription plan. If you have a Pixel 9 Pro, Pro XL, Pro Fold, or some Galaxy S25 models, you will have received a substantial free trial of this paid tier. The trial is 12 months on Pixel 9 Pro models and 6 months on S25 devices.

Google has given us a couple of ways to access the new controls. The easiest is via the dedicated Gemini app. When you launch Gemini Live, the usual call-style UI will gain an extra couple of buttons. There is a camera button and a presentation icon.

You can use your voice with the “Hey, Google” wake phrase. Above the compact Gemini pop-up bar you’ll see a “Share screen with Live” tappable button that will start AI-powered screen sharing. You are not able to select a single app yet. Instead, you have to share your entire screen. This might be intrusive for some people. A status bar chip will indicate that this is active. Tapping allows you to close or end your session quickly.

What’s it like to use?

The initial start-up and introduction is very fast, but the viewfinder opens and is very easy to decipher. It somewhat mimics the Pixel camera UI, so it’s instantly familiar to someone who has used Google phones for a while. You don’t have to use the rear camera. There is a toggle to switch to the selfie camera if you want to put yourself or your background in the frame and ask questions or advice.

When using the Gemini Live Astra mode, I have found that focus in the camera is a little finicky, as lenses switching automatically can be annoying up close. It advises you to stay still or keep the subject as still as possible. I can attest to this, but despite some subjects not being in focus, Gemini is quick to determine what something is or produce information without too much difficulty.

Not being able to zoom in, even with dedicated telephoto lenses on your phone also feels shortsighted. If you want to point out something in the world, you’re going to have to get very close up, or you hope that Gemini can work out what you’re pointing at.

For simple or basic queries, it’s perfectly adequate. However, the best use cases are for deeper questions like help or advice on objects, areas, and locations rather than asking simple things you probably could search for yourself.

A prime example is getting the calorific date on food items or allergen information. I even tried some real-time translations, but while they seemed to be fairly accurate, Gemini Live does not give or show on-screen text. You only get audio cues and answers. Google Lens overlays the translated text, so it may be a better option for translating signs, text, and more.

At the moment, you have to wait until you close a session before you can “see” some of the text-based responses. So you do get a chat log of what was said, what Gemini responded and any actions you can take. This could be a little better implemented, but for now it’s fine. In tandem with other apps like Google Maps and various other services, you could probably use this as a visual learning aid or tutorial option.

I tested Gemini Live’s Astra-style video functions on the Pixel 9 Pro XL, and although it wasn’t always instant, it felt pretty smooth, and responses were almost always fast enough to feel fluid.

Screen sharing feels a little less “natural” than using a camera. Because you get no visual feedback, merely audio cues and information, it doesn’t feel integrated in the same way. Getting webpage summaries is about the extent of the feature’s usability or getting further information without leaving a page or screen. I’d wager this will change as we get more integrations. Think the ability to add things to shopping lists with URLs or flight information to your calendar when making travel plans. We can’t do any of that yet.

Accuracy and other considerations

Like almost all AI platforms, Gemini is not a perfect system. In fact, it can get things wrong in lots of scenarios. Where it does well is with information recall.

I found that certain items caused issues where Gemini wasn’t able to give me accurate information. This is going to be fine for things you know about, and you can effectively interrupt and course-correct the wayward AI. The problem here is that hallucinations about things you might not know all that well could lead to problems.

For instance, I asked about a Gameboy game cartridge, and Gemini misidentified the cart and gave me wildly incorrect information about the title and gameplay. Making sure you have a clear view of something within the viewfinder helps to mitigate this. Just adjusting where I “filmed” instantly resolved the problem without intervention. You simply can’t trust the information blindly at this stage. Gemini is still prone to errors, and they range from minor to major in equal measure.

The world-facing camera…with caveats

Android XR and the AR platform will rely heavily on Gemini as the means to interact with and get information about the world around us. Gemini Live utilizing Astra functions is the first step in realizing that end goal. Per the demos shown late last year, we might be a few years away from Google-made AR glasses, but this early introduction isn’t a bad way to kick things off.

Like any AI product, be careful putting 100% of your trust in the information spewed out. That said, this seems like a great start and a solid way to interact with the world around you or help you when you get stuck. It’ll improve over time and get better as more data points are introduced, so at least in theory, it’s the worst that Gemini Live video modes will ever be – which isn’t all that bad to begin with.

Sharing your screen is very limited at this stage, and while it is a nice secondary option, it is even more limited. If it can develop to play nicely with more of your other services, it’ll be a useful tool. Right now, it’s a parlor trick that does little more than regular Gemini Live.

Add 9to5Google to your Google News feed.