Animating virtual avatars with free-view control through implicit Neural Radiance Field rendering (NeRF) has attracted considerable attention. Previous methods for generalizable neural human rendering employed explicit constraints to improve both quality and functional accuracy. However, directly optimizing coordinates on a complex surface leads to a dynamic semantic contradiction between the character’s pose and explicit constraints, which in turn reduces the generalizability of neural rendering for human motion. Tackling these issues, we present a novel framework named DPhuman, which optimizes the pre-fitted SMPL with a concise and consistent Hypergraph representation, integrating point registration and forward deformation into a unified model of shared rigid motion while simultaneously capturing the global topological structure. Specifically, the Hypergraph representation simplifies complex human meshes and establishes associations with human joints. Then, the Mapping-based Deformable Radiance Fields (MDRF) translate human motion into rigid translation through point registration, enabling human deformation based on specific semantic parts while mapping points. Finally, the Fine-grained module is employed to further improve fine-grained consistency, weighted from the aligned SMPL model. Extensive experiments demonstrate the superiority of our proposed DPhuman over state-of-the-art methods, and the ablation study illustrates the effectiveness of our approach.