This presentation traces the evolution of real-time avatar technology, moving beyond simple template-driven copying of a source video's movements and expressions. We will showcase the transition to explicit emotion modelling, where text prompts directly condition the avatar's facial expressions. The journey culminates in full motion modelling, demonstrating how detailed descriptions of desired movements, emotions, and actions can be used to render fully controllable, dynamic, and dancing avatars in real time.
This presentation traces the evolution of real-time avatar technology, moving beyond simple template-driven copying of a source video's movements and expressions. We will showcase the transition to explicit emotion modelling, where text prompts directly condition the avatar's facial expressions. The journey culminates in full motion modelling, demonstrating how detailed descriptions of desired movements, emotions, and actions can be used to render fully controllable, dynamic, and dancing avatars in real time.