Griffin: The First Human Interaction Model (www.tavus.io)

Today, Tavus has unveiled Griffin, the first-ever Human Interaction Model (HIM), capable of understanding and generating real-time, face-to-face human interactions. Unlike previous models, Griffin effectively listens while speaking, capturing not just words but also expressions and pauses, which significantly enhances the quality of human-machine communication. Remarkably, Griffin has passed the real-time video Turing test, with 48% of participants mistaking it for a human conversant, a major leap from the 2% pass rate of earlier systems. Griffin operates as a full-duplex video-to-video model that continuously assesses conversational context and user behavior, allowing it to react in real-time with natural emotional responses and gestures. Its architecture includes a Continuous Conversational Modeling engine and an Audio-Visual Generation engine, enabling seamless interaction that mimics human conversational flow. With rapid audio and video generation, Griffin can produce high-quality outputs while simultaneously interpreting non-verbal cues and maintaining the context of the conversation. This innovative approach represents a significant advancement in AI/ML, moving towards a future where machines communicate as naturally as humans do, potentially transforming various applications in human-computer interaction.