Artificial intelligence is moving beyond text-based chatbots with a new model designed to participate in real-time video conversations. San Francisco-based AI company Tavus introduced Griffin, which it describes as its first Human Interaction Model (HIM).
Unlike conventional chatbots, Griffin is built to communicate through video. It can listen to spoken questions, interpret visual signals, and respond with speech, facial expressions, gestures, and body movements. The company says the system is designed to make conversations feel more natural and responsive.
Tavus says Griffin does not wait until the user is done. Instead, it keeps running and tracks what is happening during the chat as it goes.
That way, when something changes mid-conversation, the system can react. It can handle people interrupting, shifts in tone, and cues from what is shown on the screen. Tavus also notes that Griffin brings together tools for making video, handling speech, and running conversational AI. The goal is to read signals like facial expressions, timing, and tone. At the same time, it keeps the talk moving in real time.
Tavus demonstrated the model in a video where Griffin responds to questions while displaying human-like movements. The company claims these capabilities make it harder for users to distinguish the AI from a person during some conversations.
Tavus made several performance claims about Griffin. The company said 48% of people who interacted with the model live believed they were speaking to a real human.
Tavus also claimed Griffin scored 3.83 out of 5 on NVIDIA’s VideoFDB benchmark. The human reference score was 3.92, while the company cited 2.80 for the next-highest published system.
These figures come from Tavus and have not been independently verified in the source report.
The technology could have applications in areas where natural conversation matters. Tavus pointed to use cases including interview practice, tutoring,a and customer support.
The model's ability to process speech and visual signals could also make it useful for interactive training and simulations. However, using an AI to impersonate a candidate in an actual job interview would raise serious questions around disclosure, consent and assessment integrity.
Some employers already prohibit candidates from using generative AI during live interviews.
Tavus did not release Griffin for unrestricted public access. The company says it is working on disclosure and safety measures before wider availability.
For now, eligible testers can access Griffin-Lite as part of the company's early testing programme. The development nevertheless signals a shift towards AI systems that do not simply answer questions but participate in more human-like, two-way video interactions.