Santiago Valdarrama

The model behind these characters is A1, It's a real-time model that creates full-body AI characters that keep listening while they talk and move. You can interrupt one of these characters mid-sentence and change the topic, and they will follow you. I read their technical report, and the secret behind how A1 works is that it generates audio and video in small, continuous segments. They keep neighboring segments consistent and, that way, reduce the drift in the character and scene over time.