Rendering · Aug 2026
Beyond the talking head: early notes on full-body presence
Some conversations are physical. Teaching yoga, rehearsing a presentation, demonstrating a stretch. Notes from building Aarohi-1.
Most conversational video stops at the shoulders. That's fine for an interview, but useless for a yoga teacher who needs to show a pose, or a presentation coach who needs to model posture.
Aarohi-1 drives a full-body rig from the same conversational state as the face: gestures are planned from the language model's output, posture shifts with conversational role, and demonstrations play from a library of motion-captured sequences.
The hard part isn't animation quality; it's coordination. A gesture that lands half a second late reads as robotic. We now plan gestures against predicted speech timing rather than reacting to it.
Aarohi-1 is in limited preview. We'll share more as it opens up.