HeadStudio: Text to Animatable Head Avatars with 3D Gaussian Splatting

• DOI: 10.48550/arxiv.2402.06149
• Publication Date: 2024-02-08
• Authors: Zhenglin Zhou; Fan Ma; Hehe Fan; Yi Yang
• Keywords: FOS: Computer and information sciences; Computer Vision and Pattern Recognition (cs.CV); Computer Science - Computer Vision and Pattern Recognition

Creating digital avatars from textual prompts has long been a desirable yet challenging task. Despite the promising outcomes obtained through 2D diffusion priors in recent works, current methods face challenges achieving high-quality and animated effectively. In this paper, we present $\textbf{HeadStudio}$, novel framework that utilizes 3D Gaussian splatting to generate realistic text prompts. Our method drives Gaussians semantically create flexible achievable appearance intermediate FLAME representation. Specifically, incorporate into both representation score distillation: 1) FLAME-based splatting, driving points by rigging each point mesh. 2) distillation sampling, utilizing fine-grained control signal guide prompt. Extensive experiments demonstrate efficacy of HeadStudio generating animatable prompts, exhibiting visually appealing appearances. The are capable rendering real-time ($\geq 40$ fps) views at resolution 1024. They can be smoothly controlled real-world speech video. We hope advance avatar creation widely applied across various domains.