Hey all! I’m trying to get the Live API quick start working. It works fine conversationally, but it doesn’t seem to understand what it’s seeing on the camera. The responses are all quite random. It seems quite similar to this previously closed issue. I’ve tried playing around by changing different methods and models, but it seems to break it even worse by throwing exceptions. Anyone have any ideas?
Hello,
Welcome to the Forum!
We recommend trying out Gemini 2.5 Flash Live to check if the issue persists there as well. If the problem continues we would require some additional details specifically whether the issue occurs with the live camera, live screen, or both. A bit more elaboration on the issue would also be very helpful for us to better analyze and support you.
Thanks @Lalit_Kumar. I did try Gemini 2.5 Flash Live but it gave me an error that one of the bidi methods wasn’t allowed. It doesn’t seem like this example was set up for the 2.5 Flash Live syntax. I only tried the issue with the camera. One of the other forum posts mentioned TURN_INCLUDES_ALL_INPUT, and enabling that got the model to respond about the visual content.
However, I’m still struggling to get it to recognise the temporal context. It doesn’t seem to understand it’s looking at a video instead of a series of individual images.
Hello,
Could you please share your code for reproduction? This will help us analyze the issue more effectively.