Add media only when it improves the actual task.
1. Frame selection
Choose frames based on task need and supported API limits. Continuous high-rate upload adds cost and bandwidth while often repeating the same information. Give the user control over camera sharing.
2. Context freshness
Label when a frame was captured and avoid answering as if an old image reflects the current scene. Handle camera revocation independently from the audio session.
3. Quality measurement
Evaluate transcription understanding, correct tool use, latency and interruption behavior in quiet and noisy environments. Include refusal and no-network paths in the demonstration.
Worked scenario
The user moves the camera after capturing a machine label. The assistant should refer to the captured label rather than claiming to see the current view.
Apply it
Add an explicit snapshot action and compare task success with and without image context.
Check your understanding
You can demonstrate media consent, freshness and a measurable benefit from the camera input. Explain the decision and show evidence from your implementation or design. If you cannot demonstrate it yet, revisit the relevant section before continuing.