DreamWaltz-G

Expressive 3D Gaussian avatars from skeleton-guided 2D diffusion. Combines skeleton-guided score distillation with hybrid 3D Gaussian avatar representation built on neural implicit fields and parameterized 3D meshes. Enables real-time rendering, human video reenactment, and multi-subject scene composition.

Paper (arXiv)GitHub Project Page

Outputs 2

library

Framework for text-driven 3D Gaussian avatar generation with real-time rendering and animation capabilities.

GitHub Project Page

GitHub Repository

DreamWaltz-G: Expressive 3D Gaussian Avatars from Skeleton-Guided 2D Diffusion

paper

Proposes skeleton-guided score distillation with hybrid 3D Gaussian representation for high-quality animatable avatar generation.

Paper (arXiv)

arXiv: 2409.17145

Venue: TPAMI 2025

vision3dgenerationopen-source