In the burgeoning field of cognitive neuroscience, the quest to visualize mental processes has spurred the development of techniques that translate cerebral signals into detailed images, offering unprecedented insights into the human brain’s workings. This research aims to harness Latent Diffusion Models (LDMs) and Large Language Models (LLMs) for transforming brain activity into visual imagery by integrating semantic mapping and image reconstruction algorithms to enhance both visual accuracy and semantic precision. LDMs are probabilistic models that generate images by iteratively refining noise, while LLMs are used to capture semantic information from textual descriptions. This research not only provides a new perspective on how the brain processes visual information, but also opens up new application areas for brain-computer interface and neuroscience research. Through detailed experimental design and strict evaluation criteria, this research verifies the effectiveness of the proposed method and explores its potential applications in medical diagnosis, rehabilitation treatment, artificial intelligence, and other fields.