LiftAvatar: Kinematic-Space Completion for Expression-Controlled 3D Gaussian Avatar Animation

LiftAvatar is a new AI system that enhances 3D avatar animation by completing sparse monocular video observations in kinematic space using expression-controlled video diffusion Transformers. The technology addresses limitations in 3D Gaussian Splatting-based avatars by generating high-quality, temporally coherent facial expressions from single or multiple reference images.