Facial landmark guidance improves 3D Gaussian Splatting talking heads
September 16, 2026
A new module uses predicted facial landmarks to provide structural cues for 3D Gaussian Splatting-based talking head synthesis. This method mitigates inaccurate mouth motion and artifacts by enriching spatial points in expression-sensitive regions.
HOW THIS AFFECTS YOU
●
researcherYou can use facial-keypoint guidance to solve temporal acoustic-to-geometry mapping issues in 3DGS models.
●
designerThis enables more realistic and stable digital human avatars for immersive media and virtual communication.