Question: Is there a way to reconstruct continuous visual experiences from noninvasive brain imaging data, like fMRI scans?

If you want to use noninvasive brain imaging data like fMRI scans to reconstruct continuous visual experiences, Mind-Video is a top contender. This project creates high-quality videos from continuous fMRI brain activity, tackling big problems like time lag and fidelity, pixel and semantic guidance, and scene dynamics and coherence. The system uses a two-module architecture with masked brain modeling and spatiotemporal attention, and scores 85% on semantic metrics.

Also worth a look is Make-A-Video, an AI information retrieval system that turns text prompts into videos. Although geared primarily for text-to-video generation, its sophisticated machine learning technology can generate realistic and fantastical videos, and could be a foundation for reconstructing video from brain imaging data in the future.

If you want a broader AI foundation, DeepBrain AI lets you generate high-quality videos from text prompts with options like customizable AI avatars and multilingual AI text-to-speech. It isn't designed to handle brain imaging data, but its video generation abilities could be extended in the future.