TRIBE v2, Meta’s AI model to predict human brain activity

Meta's research team has published TRIBE v2, an artificial intelligence model designed to simulate how the human brain processes visual, auditory and linguistic stimuli. The system generates predictions of brain activity measured by functional magnetic resonance imaging (fMRI) at a notably higher resolution than its predecessors, enabling more detailed brain maps to be produced from computational simulations.

The model was trained on MRI recordings from more than 700 healthy volunteers who were exposed to a wide variety of content: images, videos, podcasts and texts. This volume of data represents a significant step forward from the team's previous model — the winner of the Algonauts 2025 challenge — which had been trained on data from just four individuals. Thanks to this larger training dataset, TRIBE v2 can generate predictions for new subjects, new languages and new tasks without requiring specific data for those situations, a capability known as zero-shot inference.

One of the main uses Meta highlights for this tool is enabling researchers to test hypotheses about brain function without needing to recruit human participants for every experiment. This could reduce the cost and time of neuroscience studies, while also opening new avenues for investigating neurological disorders. Meta also suggests that the insights derived from this type of model could contribute to the development of AI systems more aligned with the principles of the human brain.

The model, source code, research paper and an interactive demo are publicly available under a CC BY-NC licence, allowing use for non-commercial research purposes.

Key points

• TRIBE v2 predicts brain activity in response to images, sounds and text.
• Resolution 70 times higher than comparable models using fMRI.
• Trained on more than 700 volunteers, compared to 4 in the previous model.
• Generates predictions for new subjects and languages without prior data (zero-shot).
• Allows testing of brain hypotheses without requiring human participants.
• Could accelerate research into neurological disorders.
• Findings could also be applied to the development of new AI systems.
• Publicly available with code, paper and demo under CC BY-NC licence.