MIBURI: Towards Expressive Interactive Gesture Synthesis (CVPR 2026)

Real-time dialogue-driven full-body gesture and facial-expression synthesis

MIBURI teaser
LIVE DEMO OFFLINE

The interactive demo runs on a dedicated GPU that is powered on for scheduled demo windows rather than left running continuously. When it is live, this page becomes the full dashboard: talk into your microphone and MIBURI replies with speech while a SMPL-X character gestures along in real time at 25 fps.

In the meantime, the paper, project page (with video results), and the full training + inference code are all available. The demo can also be run locally on any 24 GB GPU with miburi-demo.