Summary
- Gemini Live’s live-video mode is expanding to more Androids, now rolling out to the Galaxy S25 series.
- Initial tests indicate the “live” video is snapshot-based, acting like conversational Google Lens, not continuous analysis.
- Samsung says that the feature is rolling out to all S25 series devices at “no additional cost.”
Google Gemini Live’s real-time camera feed analysis capabilities, essentially the I/O 2024-announced Project Astra, are now starting to land on more Android devices. The multimodal assistant feature, which is able to understand audio and video prompts at the same time, first started landing on select devices roughly two weeks ago.
Fast-forward to late last week, the feature began going live for some Pixel users, and it is now starting to make its presence known on Samsung’s latest flagship.
Related
Gemini Live’s real-time camera mode gets a wider release
Only for Gemini Advanced subscribers, though
Rolling out now to the Galaxy S25 series, “Real-Time Visual AI” via Gemini Live is now free for all Galaxy S25 series users, even if they don’t have a Gemini Advanced subscription. According to Samsung, for S25 users, the functionality comes “at no additional cost.” It is currently uncertain whether that’s just a sly way of saying that ‘your new device already comes with a free Gemini Advanced Promo,’ and hence the functionality comes “at no additional cost,” or if Gemini Live’s real-time camera feed analysis indeed free-to-use now for non-Advanced users.
For what it’s worth, I can see the video mode live on an account that is subscribed to Gemini Advanced. My free account, on the other hand, still hasn’t surfaced the new feature.
Although welcome, the current implementation feels like a patchwork of different technologies. In our brief time using it, the ‘live’ aspect of the video mode wasn’t truly consistent, and it felt as though the feature wasn’t constantly watching and analyzing the video feed. Instead, the feature seems to capture a snapshot of what the camera sees at the moment when a query is posed. .
For example, asking Gemini Live to count fingers held up in real-time shows that the tool doesn’t continuously process the video feed. It seemingly bases its answer on the snapshot it takes when queried. So essentially, in its current implementation, Gemini Live’s video mode might just be Google Lens with a conversational assistant attached to it.
Similarly, in other environments, like in a car, it was able to analyze that I was in a moving vehicle (passenger seat), though it made up the name of the street I was approaching. In a different example, it said that I was approaching a road that was actually nearby, but I wasn’t driving towards it. While not entirely certain, it could be that the AI tool grabs information from Google Maps when in dynamic environments like a moving vehicle.
According to Google, it intends to extend the functionality to Android 10+ devices in the near future.