Tavus previews live AI video model Griffin
Tavus Griffin model makes AI video calls feel human with real-time responses See how it boosts natural chat and why safer rollout matters before release
Tavus has previewed Griffin, a new “Human Interaction Model” designed to look, listen, speak, and respond in real time during video calls. The company says the system can react to visual and audio cues as a person would, including noticing details on a user’s screen and responding midconversation.
In tests of GriffinLite, 48% of participants reportedly believed they were speaking with a human, up from 2.4% in earlier versions. Tavus also says the model came close to real people on NVIDIA’s VideoFDB benchmark for natural video chat and outperformed other AI models in the comparison.
Tavus is currently limiting access to trusted testers while it works on safety and disclosure features before a wider release. The demo highlights both the potential uses of more lifelike AI video, such as tutoring and caregiving, and the risks of misuse in scams or deception.