Generative models in traditional Chinese medicine: Advances and applications

Generative artificial intelligence is increasingly being applied across interdisciplinary fields, enabling cross-modal content generation-spanning text, images, and molecular structures-through self-supervised learning. Confronting persistent challenges in Traditional Chinese Medicine (TCM), such as compositional complexity, ambiguous mechanisms, and unstructured knowledge systems, generative models may provide useful approaches by bridging fragmented data sources and offering scalable tools for prescription generation, molecular design, and knowledge discovery. In practice, large language models (LLMs) convert classical texts and electronic records into structured knowledge for syndrome-aware prescription generation, while fragment-aware GPT-based chemical language models expand natural-product chemical space. This review systematically explores core paradigms and recent advances in applying generative models to TCM research, focusing on key domains including TCM knowledge Q&A, prescription generation, and molecular design. We further critically analyze methodological and ethical challenges while outlining future directions in the field, such as interpretable generative models, integration with TCM knowledge graphs, and novel multimodal training strategies.