prakhar-bhartiya/meta-tribev2-social-media-content-signal

Predicts virality of short videos using a Meta TRIBE v2 brain-encoding model

Public
11 runs

TRIBE Virality

Predicts a virality score (0–100) for short videos using Meta’s TRIBE v2 brain-encoding foundation model. TRIBE v2 predicts fMRI activation across the human cortex in response to naturalistic video, audio, and speech; this model wraps those per-vertex predictions in a cortical-proxy scoring layer that summarizes them into a single headline number plus five interpretable sub-scores.

What you get

For any short video (≤60s, MP4/MOV/WebM) the model returns:

  • Headline virality score — weighted blend of the five sub-scores.
  • Five sub-scores mapped to brain regions associated with each construct:
Sub-score Weight Cortical proxy
Reward 30% Orbitofrontal cortex, ventromedial PFC
Emotion 25% Anterior insula, anterior cingulate
Attention 20% Visual cortex, auditory cortex, intraparietal sulcus
Social Relevance 15% Medial PFC, TPJ, precuneus
Novelty 10% Parahippocampal cortex, posterior cingulate
  • Per-step timeline of all five sub-scores at ~2 Hz, for plotting how predicted brain response evolves through the clip.

Scoring is deterministic — re-running the same file yields the same score.

Intended use & limitations

  • Built for short-form social/ad creative, where the dominant signal is audiovisual + spoken language. Long clips (>60s) are not supported.
  • The “virality score” is a heuristic on top of TRIBE, not a clinical or scientific measurement. It correlates predicted neural engagement with intuitions about what tends to perform on social platforms; it is not a guarantee of real-world reach, which depends on distribution, audience, timing, and platform algorithms.
  • TRIBE v2 predicts fMRI for the average subject — individual responses vary substantially. Do not use for medical, diagnostic, hiring, lending, or other high-stakes decisions about individuals.

References

  • TRIBE v2 paper: A foundation model of vision, audition and language for in-silico neuroscience
  • Model weights: facebook/tribev2 (CC-BY-NC-4.0, Meta)
Model created
Model updated