Skip to main content
TRIBE v2

TRIBE v2

aidemos.atmeta.com

Meta's open foundation model that predicts how the human brain responds to video, audio, and text — try the interactive demo, no signup.

DevelopmentOpen SourceArtificial IntelligenceData & ResearchLearning

About TRIBE v2

TRIBE v2 is Meta AI's open foundation model for predicting human brain activity in response to naturalistic stimuli (video, audio, text). It pairs V-JEPA2 (video), Wav2Vec-BERT (audio), and LLaMA 3.2 (text) encoders, was trained on 1,000+ hours of fMRI from 720 subjects, and delivers a ~70× resolution improvement and 2–3× accuracy gain over previous methods — including zero-shot prediction for individuals it has never seen. Meta released model weights, codebase, paper, and this interactive demo so researchers can run in-silico neuroscience experiments without scanner time.

CC BY-NC licensed.

Pricing model: Free

Features

  • Predicts fMRI brain responses to video, audio, and text
  • Tri-modal foundation model (V-JEPA2 + Wav2Vec-BERT + LLaMA 3.2)
  • Trained on 1,000+ hours of fMRI from 720 subjects
  • 70× resolution + 2–3× accuracy over prior methods
  • Interactive in-browser demo — upload a clip, see predictions
  • Model weights, code, and paper released (CC BY-NC)

FAQs

Common questions about TRIBE v2

What does TRIBE v2 do?
Given a clip of video, audio, or text, TRIBE v2 predicts the fMRI brain response it would trigger — at high spatial and temporal resolution.
Is it free?
Yes — the demo is free in the browser. Weights, code, and the paper are released under CC BY-NC for research use.
Who is it for?
Neuroscience researchers, cognitive scientists, and AI/ML researchers studying perception, attention, and brain–stimulus mapping.

Similar products