Traditional material discovery methods, including experimental trial-and-error and high-throughput screening, are constrained by database scalability, hindering efficient exploration of the vast chemical space. Generative artificial intelligence (AI) is revolutionizing materials science by enabling a new paradigm for the inverse design of functional materials. This paper centers on the landmark work published in Nature—the MatterGen generative model, detailing its diffusion model-based approach for achieving stable and controllable inorganic crystal material generation. MatterGen not only generates diverse and stable crystal structures across the periodic table but also facilitates conditional generation with fine-tuning for target chemical compositions, spatial symmetries, and multiple performance constraints (e.g., mechanical, electrical, and magnetic properties). By examining the technical principles, performance advantages, and experimental validation of MatterGen, this paper illustrates how generative models are transforming material design from “screening” to “creation”, while also discussing the challenges and future development trends of this technology.