Cyborg
Recorded pose sequence · generated performer · source dancer inset
XRDance turns a dancer’s movement and spoken direction into an evolving generative stage partner—performed live, on the same stage.
Recorded dancer motion guides each generated performance, carrying a shared choreography into distinct imagined scenes.
Recorded pose sequence · generated performer · source dancer inset
Recorded pose sequence · generated performer · source dancer inset
Recorded pose sequence · generated performer · source dancer inset
Recorded pose sequence · generated performer · source dancer inset
XRDance is a projected augmented-reality performance system built around exchange. Camera imagery and optional skeletal motion condition a local video generator; spoken direction can steer the appearance of what comes next.
The system keeps capture, inference and display on one workstation. Generation overlaps video playout, so dancers can keep moving while the imagined scene develops.
Camera imagery sets the scene. Optional depth-camera tracking follows one dancer’s skeletal motion.
A resident video model synthesizes new stage imagery from visual and motion guidance.
Local speech recognition updates the text prompt at generation boundaries.
The projected image becomes a changing partner for the next movement.
XRDance has been accepted for presentation in the Real-Time Live! program at ACM SIGGRAPH Asia 2026 in Kuala Lumpur, Malaysia, 1–4 December 2026.
The project’s extended abstract draft is available here. It describes the system, measured implementation and current validation limits.