A new study presents a multimodal framework that combines BLIP-2, a vision-language model, with Stable Diffusion to decode brain activity from functional magnetic resonance imaging (fMRI) data. This innovative approach translates neural signals into structured text descriptions and extracts semantic information, facilitating visual reconstruction and detailed neural feature analysis. By aligning fMRI-derived embeddings with a shared visual-textual space, the framework captures a range of information from basic perceptual features to complex semantic concepts. Experimental results indicate that this method effectively preserves essential visual attributes in neural representations. The findings highlight the framework’s potential to enhance fMRI-based multimodal decoding, offering deeper insights into the neural processes involved in visual perception. Clinically, this research could lead to tools for decoding and reconstructing mental imagery, benefiting healthcare professionals in diagnosing and treating neurological disorders. The framework may also support rehabilitation, early detection of cognitive decline, and the creation of brain-to-text interfaces for communication among patients with severe motor disabilities.