Phoenix-4 adds emotion to avatars

Tavus is pushing the avatar stack past lip-sync demos and into something closer to usable human-computer presence. If Phoenix-4 performs as advertised, the competitive line shifts from “can it talk in real time?” to “can it react like a person while listening?”

• –The key claim is not just photorealism but behavior realism: Phoenix-4 generates listening cues, micro-expressions, and emotion changes instead of looping canned footage.
• –Tavus frames Phoenix-4 as a model release inside a larger stack with Raven and Sparrow, which makes it more relevant to developers building end-to-end conversational agents than a standalone avatar demo.
• –The published benchmark claims—40 fps at 1080p with full head rendering—directly target weaknesses in competitors that still trade off resolution, latency, or expressive control.
• –This is a meaningful step for AI support, coaching, healthcare, and sales use cases where user trust depends on whether the agent feels attentive rather than merely responsive.