Tavus introduces interactive video avatar model Griffin

Video technology startup Tavus announced a conversational artificial intelligence model called Griffin that generates interactive video avatars during live video calls. Unlike older digital assistants that convert speech to text before generating a response, Griffin processes visual and audio cues simultaneously while speaking. According to testing data shared by Tavus, 48 percent of human participants believed they were conversing with a real person during a one minute video interaction.

This shift matters because human communication relies heavily on non-verbal signals like facial expressions, timing, and vocal tone. Traditional digital avatars often interrupt users, miss visual context, or create uncomfortable pauses. By handling sight and sound together in real time, synthetic avatars could make digital customer service, virtual tutoring, and remote healthcare check-ins feel far more natural.

However, several aspects of the technology remain unproven. Tavus has currently released only a early test version to a limited group of developers, meaning independent evaluations are still needed. It is unclear how the model will perform during poor internet connections, heavy accent variations, or multi-person calls. Moreover, as video avatars become harder to distinguish from actual people, regulators and consumers will face growing questions about whether mandatory disclosure should be required on automated video calls.