In this article
When you review a live video avatar, time the wait for its first reply, watch for stalls after playback starts and check whether the mouth matches the speech. Log each observation separately. A fast reply can still have a frozen frame or speech that arrives ahead of the mouth.
What Runway measured
In its November 5 research report, Runway reported a median wait of 2.13 seconds from the end of user speech to audible avatar speech across 51 turns in 12 sessions. The research paper describes producing four frames at a time for playback at 24 frames per second. Each group covers about one sixth of a second. Keeping up with that pace is a continuous job. Use the reported median as context for the sample Runway studied, and set your own review criteria around the experience your audience needs.
Time the wait from the viewer’s side
Imagine a fictional maker-fair booth with a live character that explains how a desk organizer fits together. Prepare a short question about one piece and a longer question asking for the assembly order. Repeat both on the device the booth will use. Keep the wording consistent so you can compare attempts.
In a recording of the demo, mark where the visitor finishes speaking and where the character’s first audible word begins. Note what the visitor sees while waiting. A visible listening or processing cue can make the state clearer. Log the wait and the cue separately so a designer can address each.
Watch the full answer for interruptions
Follow the answer through its last word. Mark any place where the picture freezes or sound drops. Write whether the face, voice or both stopped. This gives the team a place to look in the recording. On a longer response, note whether the problem appears as the answer begins or after several phrases.
Record the device and connection conditions with each attempt. If playback changes when you move to the actual booth network, note the difference and repeat the same questions there. Use the evidence to decide where to investigate further.