Facial proportions
Balance, thirds, fifths and visible landmark relationships
PEPRA METHODS · VERSION 1.1
A score you can inspect.
A plan you can test.
PEPRA turns one standardized selfie into visible measurements, uncertainty ranges and ranked next actions. It does not turn a photo into a diagnosis, identity judgment or promise.
Adults 18+ · contextual estimates · human-review roadmap01 · THE MODEL
Not one mystery number.
The shareable score is a transparent summary. Every category remains available underneath it with its own estimate, confidence interval, visible evidence and limitations.Symmetry
Pose-aware side-to-side relationships
Eyes + brows
Spacing, tilt, openness and brow framing
Skin
Visible evenness, texture, pigmentation and redness
Smile
Smile balance and activation from an appropriate capture
Hairline
Visible hairline shape, temple balance and density
02 · WHAT WE MEASURE
Six lenses. Twenty-three face categories.
Landmarks come from ML
Web captures use MediaPipe Face Landmarker; iPhone captures use Apple Vision. The detector name and version stay attached to every measurement record.
Geometry is deterministic
Code—not GPT—calculates supported ratios and angles from detected points. The same stored point map drives both the displayed number and the overlay.
Visible skin + hair
Evenness, texture, pigmentation, redness, hair density and hairline—kept separate rather than collapsed into “skin quality.”
GPT explains; it does not measure
The language model describes visible context and ranks reversible actions. It cannot replace or contradict deterministic geometry, and it must mark unsupported measurements unavailable.
Style Lab stays separate
Face-framing color combinations, contrast, silhouette, part, length, fringe and volume never affect the facial score. Skin tones are never ranked.
03 · FROM IMAGE TO ACTION
Every recommendation must pass five gates.
Observable
Is the feature visible in this image without guessing a sensitive trait?
Repeatable
Would a second standardized photo produce a similar estimate?
Actionable
Is there a lower-cost, lower-risk step the member can actually test?
Measurable
Can progress be compared with the same pose, light and camera?
Bounded
Are uncertainty, conflicts and professional-review requirements explicit?
04 · EVIDENCE LEDGER
Research informs the question. Your response decides the plan.
Perception research is heterogeneous and often population-specific. PEPRA cites it as context, not as a universal beauty law.REVIEW
Facial attractiveness: evolutionary based research
Reviews symmetry, averageness, color and texture while emphasizing individual and cultural variation.
Read source ↗OBSERVATIONALSkin color homogeneity and visual perception
Associates visible skin-color distribution with perceived age, health and attractiveness in a limited study population.
Read source ↗PSYCHOPHYSICSSkin homogeneity and perceived facial change
Tests how controlled changes in skin homogeneity affect attractiveness judgments and change detection.
Read source ↗CAUTIONDo averageness and symmetry signal health?
Shows that perceived health and actual health are not interchangeable; symmetry did not establish actual health.
Read source ↗TRIALRandomized photoaging formulation study
Supports testing consistent, low-risk skin routines—not predicting a personal outcome from one photo.
Read source ↗TRIALUVA1 photoprotection trial
Provides controlled evidence about photoprotection while remaining product- and population-specific.
Read source ↗05 · NON-NEGOTIABLE LIMITS
Useful, not cruel.
- No inference of race, ethnicity, sexuality or personality.
- No diagnosis or prescription from an image.
- No skin-tone ranking or skin-lightening recommendations.
- No body dysmorphia reinforcement or punitive language.
- No medical or cosmetic procedure without professional review.
- No claim that a generated preview predicts a real outcome.
- No health or fitness signal is used to calculate attractiveness.
THE PERSONAL-RESPONSE STANDARD
Keep what works. Retire what does not.
Same poseSame lightSame cameraOne variableDeclared time window
A before-and-after card should show both the model hypothesis and the member's measured response. If the response does not support the hypothesis, PEPRA visibly retires it.
Build my private report →