AI Diffusion Model Reconstructs Human Vision From fMRI Brain Scans Using Synthetic Data Loop

The Weizmann Institute team combined a dual-branch brain decoder—predicting both image structure and content—with a diffusion model to recreate visual stimuli directly from fMRI neural activity data. The system maps brain volumes down to a single cubic millimeter of neurons, based on initial high-resolution scans taken from eight subjects. To bypass the severe hardware bottleneck of acquiring human fMRI data, researchers trained a reverse AI encoder to predict synthetic brain activity from thousands of target images, creating an infinite data loop for the primary decoder. While the technology is currently restricted to clinical environments with room-sized fMRI machines, researchers caution that mapping these weights to portable consumer EEG devices could eventually allow passive neural data extraction. "If there’s a way to surreptitiously extract information about what you’re thinking about, then... 150 years of sci-fi can come true anytime, and that’s worrisome in a lot of ways." — Tommy Sprague