HyperHuman is a model for generating realistic human images. It captures the structural features of human images, ranging from coarse body skeletons to fine-grained spatial geometry, to generate human images with coherence and naturalness. HyperHuman consists of three parts: 1) Building a massive human dataset called HumanVerse, which contains 340M images and comprehensive annotations such as human poses, depth, and surface normals; 2) Proposing a latent structure diffusion model that simultaneously denoises depth, surface normals, and the synthesized RGB image. Our model forces learning of image appearance, spatial relationships, and geometry within a unified network, with each branch exhibiting structural awareness and texture richness; 3) Finally, to further enhance visual quality, we propose a structure-guided refiner for more detailed high-resolution generation. Extensive experiments demonstrate that our model has generated human images with high realism and diversity in various scenarios, achieving state-of-the-art performance.