By No Code MBA · Facts checked
- Provider
- Tavus
- Announcement date
- Status at last check
- Limited access or preview
- Model family
- Griffin
- Capabilities
- Audio, Video
What changed
Griffin is Tavus's first Human Interaction Model, a real-time video-to-video system that takes a user's audio and video and generates speech, 720p video and nonverbal behavior from a single reference image. It merges perception, turn-taking and rendering that Tavus previously split across Raven-1, Sparrow-2 and Phoenix-4.5. Tavus reports average video generation latency of 0.43 seconds on H100s, voice cloning from about 10 seconds of audio, and a 48% believed-real rate in a 54-person study versus 2.4% for Phoenix-4.5.
Access and availability
Only Griffin-Lite is available, as a research preview for a select group of early testers. It is not on the Tavus platform or API, and others can join a waitlist on the Griffin page.
Official access documentationWhat you can build
For a face-to-face agent prototype, evaluate whether live video and interruption handling improve conversation flow, and assess disclosure and consent requirements before user testing.
Limitations
Tavus says Griffin will only be released after further alignment and safety work, including disclosure features, because it could deceive people into thinking they are not talking to AI. Conversational flow was the lowest-rated area in its evaluations, and the study used one-minute calls.