Neural Radiance Field (NeRF) has gained widespread attention in the research of novel view synthesis and 3D shape reconstruction due to their powerful rendering capabilities in recent years. Although volume density of NeRF can reconstruct surface geometry, its probabilistic statistical properties often result in suboptimal surface quality. Additionally, NeRF is represented a large continuous implicit field, which limits its ability to generalize across multiple scenes and requires significant training time. To address these issues, we propose a novel network: Hyper-NeuS. Built on the Auto-Decoder framework, Hyper-NeuS employs SDF-based volume rendering to create a neural radiance field, leveraging Hypernetworks to learn shape priors and refine parameters of the neural radiance field. Finally, the SDF isosurface extraction algorithm is used to generate the surface geometry. Experimental results demonstrate that our model significantly improves surface reconstruction speed and quality over state-of-the-art methods while greatly enhancing the generalization capability of neural radiance fields.