Face Explorer 🔬

532 faces from the aella.lol face games · auto-measured geometry × crowd-perceived personality × attractiveness by rater sex · companion to the write-up · curated rating-set photos only (no user submissions) · includes retired photos that earned enough data before leaving the games
click any face — a scatter point, a matrix cell's photos, or a thumbnail — to pin it here
Reading this honestly: r values carry 95% CIs — with ~150 faces per set, |r| under ~.16 is not distinguishable from zero. Perceived-trait means need ≥5 raters (median ~20) — rater noise attenuates correlations, so true effects are modestly larger. Turned/tilted faces are excluded from geometric axes by default (pose ruins the ratios). "Control smiling + apparent age" shows partial correlations — geometry net of expression and age. Win rates come from head-to-head matchups (≥30 per rater sex per face); matchmaking compresses their spread. Rater sex = AMAB/AFAB self-report. Retired faces (left the games in curation passes) keep their earned data and are always included. Paired a/b photos of the same person contribute data but their images are withheld; group comparisons treat photos, not people, as the unit for pairs. Everything here is a per-photo aggregate — no individual respondent data. Data: 2026-08-03 · the write-up · rate faces yourself at aella.lol.