AI Reconstructs 'High-Quality' Video Directly from Brain Readings in Study

Researchers from the National University of Singapore and The Chinese University of Hong Kong, including Jiaxin Qing, Zijiao Chen and Juan Helen Zhou, have built MinD-Video, a model that uses fMRI brain readings and a fine-tuned version of the Stable Diffusion image model to generate video. The paper was posted to the arXiv preprint server last week. The researchers trained the system on a publicly available dataset of fMRI readings from subjects watching videos, and they describe the reconstructed videos as "high-quality" in motion and scene dynamics. They also report an accuracy of 85 percent, an improvement on previous approaches.