Summary
Tavus has released Griffin-Lite to a limited group of research testers, previewing a real-time video-to-video system that watches, listens, speaks and generates an expressive scene during a call. The company describes Griffin as a Human Interaction Model combining continuous audiovisual perception, conversational decision-making and streaming generation. Tavus says the system can respond to visual context, manage interruptions and create a full scene from a reference image. In a company-run study, 26 of 54 participants believed a one-minute Griffin call was with another person. That 48% result is not an independent certification, and Tavus has not announced general availability. The company says it is restricting access while it develops disclosure and safety measures, acknowledging that lifelike live video can weaken appearance and mannerisms as evidence of identity.
Hana’s perspective
I think the creative opportunity is obvious: live characters that can see, listen and improvise could make interactive animation and virtual performers feel less like scripted chatbots. The identity risk is equally central. A one-minute company test is too small to establish general human equivalence, yet 48% confusion is enough to show why visual realism cannot serve as authentication. Any production using this technology should make synthetic identity persistent and unmistakable, while sensitive services should verify people through channels independent of the video feed.
Are you a writer with a great story? We'd love to hear from you: APPLY HERE