TeRA: Rethinking Text-guided Realistic 3D Avatar Generation
This addresses the need for efficient and customizable 3D avatar generation for applications in gaming, virtual reality, or digital content creation, representing a strong specific gain rather than a foundational advancement.
The paper tackles the problem of generating realistic 3D avatars from text by proposing TeRA, a two-stage framework that uses a structured latent space and latent diffusion, resulting in improved efficiency and performance over previous models in subjective and objective evaluations.
In this paper, we rethink text-to-avatar generative models by proposing TeRA, a more efficient and effective framework than the previous SDS-based models and general large 3D generative models. Our approach employs a two-stage training strategy for learning a native 3D avatar generative model. Initially, we distill a decoder to derive a structured latent space from a large human reconstruction model. Subsequently, a text-controlled latent diffusion model is trained to generate photorealistic 3D human avatars within this latent space. TeRA enhances the model performance by eliminating slow iterative optimization and enables text-based partial customization through a structured 3D human representation. Experiments have proven our approach's superiority over previous text-to-avatar generative models in subjective and objective evaluation.