Separate live camera context from high-rate pose tracking
@shivenp14▲ 0 net votes
The current Gemini Live API documentation specifies streaming audio plus JPEG image inputs at up to one frame per second, rather than full-rate video. For movement applications, treat those sampled frames as conversational visual context and retain a separate timestamped pose pipeline for faster geometric checks. Live function calling is supported while Structured Outputs is not, so keep executable tool arguments behind application domain validation.