AURA — Real-time facial analysis (YouTube)

On-device facial analysis for YouTube: FACS Action Units, expressions and microexpressions. Not a lie detector.

AURA overlays real-time facial analysis on any YouTube video — and is honest about the one thing it cannot do. WHAT IT MEASURES • Action Units (FACS) — ~20 AUs estimated per frame from Google MediaPipe's 52 blendshapes • Expressions — EMFACS prototypes (happiness, anger, fear, surprise, sadness, disgust, contempt) • Microexpressions — rapid AU spikes (onset + offset under ~600 ms), logged live • A facial-landmark mesh painted over the player 100% ON-DEVICE PRIVACY All analysis runs locally in your browser through MediaPipe + WebAssembly. No video frame is ever uploaded, stored, or sent to any server. There is no account, no tracking, no telemetry. HONEST BY DESIGN This is NOT a lie detector. Detecting deception from the face is scientifically unreliable — humans score about 54% (barely above chance; Bond & DePaulo, 2006), and large behavior-detection programs built on the idea have failed audit (TSA SPOT; GAO, 2013). The optional "incongruence index" is clearly labeled as observational and non-diagnostic. Treat every reading as an estimate, not a verdict, and never use it to judge, screen, or accuse anyone. HOW TO USE 1. Open any youtube.com/watch video with a visible face 2. The analysis panel appears on the right — toggle it from the toolbar icon Built on Google MediaPipe FaceLandmarker.