ABSTRACT Generative models are redefining alloy design by moving beyond property prediction toward the autonomous creation of new compositions, processing parameters, microstructures, and architectures. Unlike conventional machine learning methods that map material descriptors to properties, generative frameworks learn the underlying probability distributions across composition, processing, and microstructure. This capability enables efficient exploration of vast design spaces while reducing the risk of local optimization. This review establishes a unified framework linking metallurgical objectives with generative modeling tasks, encompassing property optimization, inverse design for target properties, and microstructure or architecture generation. We detail how generative models address these tasks by outlining methodological foundations, highlighting representative case studies, and assessing both their strengths and limitations. Key challenges, including data scarcity, experimental uncertainty, and limited interpretability, are discussed alongside emerging opportunities in optimization‐driven workflows, active learning, and automated experimentation. Together, these advances position generative modeling as a cornerstone of accelerated and autonomous alloy discovery.
Cheng et al. (Sat,) studied this question.