// What it does
Upload two versions of any content — ad copy or images —
and get back a prediction of how the human brain would respond to each one.
Built on Meta's open-source TRIBE v2 neural encoding model, the simulator predicts realistic
fMRI-style brain activity patterns for each stimulus. It then compares them across key cortical
regions — visual, auditory, language, attention, limbic — and declares a winner with
confidence scores, activation graphs, and brain surface maps.
Think of it as an A/B test where the measurement instrument is a simulated brain, not a click-through rate.
Useful for marketers, content teams, UX designers, and researchers who want directional,
data-grounded decisions before investing in production or live testing.
TRIBE v2
Meta AI
Python 3.10+
Gradio
PyTorch
Plotly
nilearn
scikit-learn
// How it works
01
Upload your variants
Add Variant A and Variant B — video, audio, image, text, or any multimodal combination.
Each variant gets its own modality setting.
02
TRIBE predicts brain activity
The TRIBE v2 model analyzes each stimulus and generates a predicted fMRI activation
pattern across 20,484 cortical vertices over time.
03
Neural responses compared
Welch's t-test + Cohen's d + bootstrap CI on time-aligned crops. Region-by-region
breakdown across visual, auditory, language, and attention networks.
04
Winner declared
Confidence-rated result (High / Medium / Low) with proxy scores for Engagement,
Attention, and Emotion — plus a full CSV export.
// What you get
🏆
Winner
Statistically grounded result with High / Medium / Low confidence rating
📈
Activation timeline
Time-aligned mean activation curves for both variants on a shared axis
🧠
Brain surface maps
fsaverage5 cortical surface plots showing peak activation per vertex
📊
Region breakdown
12 cortical regions compared — visual, auditory, prefrontal, limbic, and more
🎯
Proxy scores
Engagement, Attention, and Emotion on a 0–100 scale per variant
📥
CSV export
Full results table with per-region activations, stats, and metadata
// Try it live