Brief

Researchers have developed Edit3DGS, a novel framework that unifies 2D instruction-guided diffusion models with 3D Gaussian splatting for dynamic 3D head editing. This approach allows for semantic control over facial regions using text prompts, enabling modifications like expression changes and attribute refinements. The edited frames are then synthesized into a coherent, high-fidelity 3D avatar that maintains identity and motion, with applications in virtual avatars and media production. AI