Reconstructed video of what a mouse saw from brain activity

A team at University College London showed mice video clips and recorded cells in the visual cortex one by one. Then they built a dynamic model: how a particular neuron should behave on a blank screen and during a film. After that, they adjusted the pixels of the reconstruction until the predicted activity matched the real activity.

On a new 10-second clip that the network hadn’t seen, the image still came together. The more cells were included in the recording, the closer the match became.

There is pixel-level agreement, but with a small shift in time. Sharpness is moderate, the field of view is incomplete. This isn’t “reading a person’s thoughts” and not a ready-made dream decoder. It’s a tool: you can compare what was on the screen with what the cortex made of it.

The authors emphasize exactly that. The brain isn’t making a mistake when it slightly alters reality. It’s interpreting. Reconstruction makes it possible to see that edit directly — not from a button the mouse pressed, but from the very frame the neurons agreed to show.

Sources: University College London. "Scientists recreated what mice saw from brain activity alone." ScienceDaily (16 September 2026).

Bauer J., Margrie T.W., Clopath C. "Movie reconstruction from mouse visual cortex activity." eLife (2026). DOI: 10.7554/eLife.105081.3.

Illustration: Sciencedaily