Trimodal brain encoder · 720 subjects · 1,115h fMRI

Predict virality by simulating the brain that watches.

Upload a video or paste a URL. We map it onto a digital twin of human cortex — 70,000+ voxels — then forecast engagement using the same anticipatory affect signatures that predicted YouTube view counts in Stanford's neuroforecasting studies.

YouTube and Vimeo URLs render a thumbnail-driven analysis. For full frame-by-frame neural mapping, upload an .mp4 / .webm / .mov file.

70k+ cortical voxels modeled

9 brain networks tracked

r = 0.65 NAcc → view-count correlation*

TL;DR · Plain-English readout —

Calculating…

Score — / 100

Percentile — vs. archetype library

Closest archetype — —

Biggest lever — —

—

Virality Forecast

0 / 100

Calculating…

Reward (NAcc)

—

Aversion (AIns)

—

Value (MPFC)

—

Attention (PCC)

—

Untitled stimulus —

Cortical activation map

Predicted BOLD response averaged across the stimulus. Hotter = higher voxel activation. Hover regions for details.

coolbaselinehot

Engagement timeline

Per-second activation across reward, aversion, and value-integration networks. The first ~4 seconds carry the strongest signal in real neuroforecasting studies.

Reward (NAcc) Aversion (AIns) Value integration (MPFC)

Benchmark vs. viral archetypes

Where this video falls compared to neural fingerprints we built from 12 reference categories of high-view content.

What the brain is telling us

Evidence-aligned readouts of where the response is strong or weak.

Tips & recommendations

Prioritized edits to lift your score. Estimated point lifts are directional, based on the same neural levers that drove the Stanford NAcc/AIns regression on YouTube view counts.

Headroom — pts available

Estimates assume one change at a time; combined gains rarely add linearly. Re-scan after editing to measure actual lift.

How the score is built

We extract trimodal features from your video — visual frames at 1 Hz, audio energy and spectral character via Web Audio, and any caption / OCR text — then map them to nine canonical brain networks defined by Meta's TRIBE v2 atlas: V1, FFA, EBA, PPA, STS, language (Broca/Wernicke), MPFC/PCC, NAcc, and AIns.

The virality score follows the Knutson lab AIM framework: positive arousal at video onset (NAcc↑) and low negative-affect signal (AIns↓) are the only neural measures that scaled out-of-sample to YouTube view counts. Onset weighting is set to 1.6× to mirror the first-4-second effect they reported.

Disclaimer. This is an in-silico simulation, not a clinical fMRI scan. Activation values are derived from interpretable proxies of TRIBE v2's published feature behaviour, not the model's released weights. For research and creative directional use only.

Predict virality by simulating the brain that watches.

Encoding stimulus into voxel space…