Human Avatar Animation

Human avatar animation studies how to create digital people whose appearance and motion remain coherent as they speak, gesture, change pose, or interact. Systems reconstruct avatars from images or video using meshes and parametric models such as FLAME and SMPL-X, neural radiance fields, or 3D Gaussian Splatting, then animate them with pose retargeting, learned deformation, or physical simulation. Diffusion Transformers and flow-matching models increasingly generate motion from audio, video references, or semantic controls, with evaluations covering identity, lip synchronization, expression, and temporal consistency. A central challenge is balancing realistic, varied motion and generalization to new people with identity fidelity, stable long-form behavior, and real-time performance.