While Pixel 9 and Samsung Galaxy S25 owners have had access to Gemini Live’s camera mode for a while now, during its I/O conference earlier this month, Google announced that the feature started its rollout for all Android users and iOS users, too. The big news here is that iPhone owners can now have access to one of the coolest AI features we’ve seen in a while now, especially since that all other Android users supposedly got access to the camera mode back in April. 

If you’re unaware of what the camera mode feature is, to put it in simple terms, Google successfully gave Gemini the ability to see, as it can recognize objects that you put in front of your camera. 

It’s not just a party trick, either. Not only can it identify objects, but you can also ask questions about them — and it works pretty well for the most part. In addition, you can share your screen with Gemini so it can identify things you surface on your phone’s display. When you start a live session with Gemini, you now have the option to enable a live camera view, where you can talk to the chatbot and ask it about anything the camera sees. 

I spent some time with it when it showed up on my Pixel 9 Pro XL in early April and was pretty wowed overall. I was most impressed when I asked Gemini where I misplaced my scissors during one of my initial tests.

«I just spotted your scissors on the table, right next to the green package of pistachios. Do you see them?»

Gemini Live’s chatty new camera feature was right. My scissors were exactly where it said they were, and all I did was pass my camera in front of them at some point during a 15-minute live session of me giving the AI chatbot a tour of my apartment.

When the new camera feature popped up on my phone, I didn’t hesitate to try it out. In one of my longer tests, I turned it on and started walking through my apartment, asking Gemini what it saw. It identified some fruit, ChapStick and a few other everyday items with no problem. I was wowed when it found my scissors. 

That’s because I hadn’t mentioned the scissors at all. Gemini had silently identified them somewhere along the way and then  recalled the location with precision. It felt so much like the future, I had to do further testing. 

My experiment with Gemini Live’s camera feature was following the lead of the demo that Google did last summer when it first showed off these live video AI capabilities. Gemini reminded the person giving the demo where they’d left their glasses, and it seemed too good to be true. But as I discovered, it was very true indeed.

Gemini Live will recognize a whole lot more than household odds and ends. Google says it’ll help you navigate a crowded train station or figure out the filling of a pastry. It can give you deeper information about artwork, like where an object originated and whether it was a limited edition piece.

It’s more than just a souped-up Google Lens. You talk with it, and it talks to you. I didn’t need to speak to Gemini in any particular way — it was as casual as any conversation. Way better than talking with the old Google Assistant that the company is quickly phasing out.

Google also released a new YouTube video for the April 2025 Pixel Drop showcasing the feature, and there’s now a dedicated page on the Google Store for it.

To get started, you can go live with Gemini, enable the camera and start talking. That’s it.

Gemini Live follows on from Google’s Project Astra, first revealed last year as possibly the company’s biggest «we’re in the future» feature, an experimental next step for generative AI capabilities, beyond your simply typing or even speaking prompts into a chatbot like ChatGPT, Claude or Gemini. It comes as AI companies continue to dramatically increase the skills of AI tools, from video generation to raw processing power. Similar to Gemini Live, there’s Apple’s Visual Intelligence, which the iPhone maker released in a beta form late last year. 

My big takeaway is that a feature like Gemini Live has the potential to change how we interact with the world around us, melding our digital and physical worlds together just by holding your camera in front of almost anything.

I put Gemini Live to a real test

The first time I tried it, Gemini was shockingly accurate when I placed a very specific gaming collectible of a stuffed rabbit in my camera’s view. The second time, I showed it to a friend in an art gallery. It identified the tortoise on a cross (don’t ask me) and immediately identified and translated the kanji right next to the tortoise, giving both of us chills and leaving us more than a little creeped out. In a good way, I think.

I got to thinking about how I could stress-test the feature. I tried to screen-record it in action, but it consistently fell apart at that task. And what if I went off the beaten path with it? I’m a huge fan of the horror genre — movies, TV shows, video games — and have countless collectibles, trinkets and what have you. How well would it do with more obscure stuff — like my horror-themed collectibles?

First, let me say that Gemini can be both absolutely incredible and ridiculously frustrating in the same round of questions. I had roughly 11 objects that I was asking Gemini to identify, and it would sometimes get worse the longer the live session ran, so I had to limit sessions to only one or two objects. My guess is that Gemini attempted to use contextual information from previously identified objects to guess new objects put in front of it, which sort of makes sense, but ultimately, neither I nor it benefited from this.

Sometimes, Gemini was just on point, easily landing the correct answers with no fuss or confusion, but this tended to happen with more recent or popular objects. For example, I was surprised when it immediately guessed one of my test objects was not only from Destiny 2, but was a limited edition from a seasonal event from last year. 

At other times, Gemini would be way off the mark, and I would need to give it more hints to get into the ballpark of the right answer. And sometimes, it seemed as though Gemini was taking context from my previous live sessions to come up with answers, identifying multiple objects as coming from Silent Hill when they were not. I have a display case dedicated to the game series, so I could see why it would want to dip into that territory quickly.

Gemini can get full-on bugged out at times. On more than one occasion, Gemini misidentified one of the items as a made-up character from the unreleased Silent Hill: f game, clearly merging pieces of different titles into something that never was. The other consistent bug I experienced was when Gemini would produce an incorrect answer, and I would correct it and hint closer at the answer — or straight up give it the answer, only to have it repeat the incorrect answer as if it was a new guess. When that happened, I would close the session and start a new one, which wasn’t always helpful.

One trick I found was that some conversations did better than others. If I scrolled through my Gemini conversation list, tapped an old chat that had gotten a specific item correct, and then went live again from that chat, it would be able to identify the items without issue. While that’s not necessarily surprising, it was interesting to see that some conversations worked better than others, even if you used the same language. 

Google didn’t respond to my requests for more information on how Gemini Live works.

I wanted Gemini to successfully answer my sometimes highly specific questions, so I provided plenty of hints to get there. The nudges were often helpful, but not always. Below are a series of objects I tried to get Gemini to identify and provide information about.