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.

• →LiftAvatar uses kinematic space completion to improve 3D avatar animation quality from sparse monocular video inputs.
• →The system combines shading maps with expression coefficients for precise and stable facial animation driving.
• →Multi-reference conditioning enables strong 3D consistency by aggregating complementary cues from multiple frames.
• →The technology serves as a plug-and-play enhancer for existing 3D Gaussian Splatting-based avatar systems.
• →Extensive testing shows consistent improvements in animation quality, especially for extreme and unseen expressions.