Tavus introduced Griffin, an AI model for real-time face-to-face interaction, while restricting Griffin-Lite to selected trusted testers. The system processes sound and video as it generates speech, expressions and movement. It can respond to gaze and pauses while a conversation continues. Tavus says customer access awaits further safety work, including features that disclose the model’s AI identity before a broader release.
Tavus, a company developing conversational video AI, introduced Griffin on 1 October for real-time face-to-face interaction. Griffin-Lite is offered as a research preview restricted to selected trusted testers. Its design processes incoming sound and video while generating the face and voice of a conversational partner. Customer access remains on hold while Tavus works on additional safety and identity-disclosure features.[1], [2]
A continuous conversational engine processes audio and video and decides what to say and how to behave. An audiovisual generator turns those decisions into streaming speech and video. Tavus describes reactions to gaze, expressions, pauses and spoken context, with decisions made during the exchange rather than only at the end of a speaker’s turn.[1]
Tavus says one reference image supplies the starting point for generating the whole scene. The output includes body movement, the chair, shadows and the background, as well as the face. The company acknowledges that a more convincing conversational partner brings deception risks. It says disclosure features are under development before a broader release, keeping the research preview separate from the existing customer platform.[1]
