Meta researchers develop TRIBE v2 to predict brain responses to video, audio, and text

Meta-affiliated researchers published TRIBE v2, a model that predicts brain responses to video, audio, and text.

TRIBE v2 was trained on fMRI data from 720 participants totaling over 1,000 hours and reproduces known localized brain activations.

The model outperformed linear approaches, won Algonauts 2025, and can run in-silico experiments without new fMRI scans.

Limitations include fMRI temporal resolution, exclusion of smell and touch, and no modeling of action or decision processes.