Pretty Scale — Face Symmetry & Beauty Score Analyzer
Facial symmetry, golden-ratio proportions, and feature balance — scored by AI from 468 facial landmarks. Runs entirely in your browser, no upload required.
Upload a clear, front-facing photo
Drag & drop or click to browse
Your photo stays on your device — it is never uploaded to any server.
Warming up AI model in the background…
How Pretty Scale works
Three-stage AI pipeline running entirely on your device.
Upload a photo
Select any front-facing photo. The AI works on JPEG, PNG, or WEBP. A high-resolution, well-lit photo gives the most accurate landmark positions.
468 landmarks mapped
MediaPipe FaceLandmarker plots 468 precise points across your face — eye corners, brow peaks, nose contour, lip borders, jaw edge, and more.
Geometry scored
The scoring algorithm computes symmetry ratios, golden-ratio deviation, canthal tilt, and facial thirds balance from the landmark coordinates.
What the score measures
Six geometric dimensions combined into a single 1–10 score.
Facial symmetry
The degree to which the left and right halves of your face mirror each other. Symmetry is one of the most consistent cross-cultural signals of health and attractiveness.
Golden ratio proportions
The phi ratio (≈1.618) appears repeatedly in idealised facial proportions — eye width to face width, nose-to-chin distance, and more. Closer measurements approach higher scores.
Canthal tilt
The angle of the outer eye corners relative to the inner. An upward tilt is associated with a youthful, lifted eye appearance, which contributes positively to the score.
Facial thirds
Classic beauty standards divide the face into equal thirds: hairline-to-brow, brow-to-nose-base, nose-base-to-chin. Balanced thirds indicate harmonious vertical proportions.
Inter-feature spacing
The relative spacing between eyes, nose, and mouth. The tool measures how these intervals compare to idealised geometric relationships mapped across annotated face data.
Feature prominence
Jawline definition, brow prominence, and cheekbone position contribute geometric signals measured through landmark distances, not subjective appearance judgments.
Understanding your score
100% private — no upload
MediaPipe FaceLandmarker runs in WebAssembly directly in your browser. Your photo never leaves your device and is removed from memory the moment you refresh or close the page.
Tips for the most accurate score
- Front-facing photo — head turned even slightly changes landmark positions
- Neutral expression — smiling stretches features and skews measured ratios
- Good, even lighting — avoid shadows across one side of the face
- Hair away from the face so jaw and brow landmarks are fully visible
- No glasses or large accessories near the face boundary
- High-resolution, in-focus image for the most precise landmark detection
Pretty Scale — frequently asked questions
What does the Pretty Scale score actually measure?
The score is a composite of geometric measurements: bilateral facial symmetry, how closely your proportions approach the golden ratio, canthal tilt, facial thirds balance, and inter-feature spacing. It is not an aesthetic judgment — it is a mathematical score based on landmark geometry.
Does a higher score mean more attractive?
Not necessarily. The score measures geometric idealisation, not human attractiveness in the full sense. Attractiveness is influenced by expression, charisma, confidence, grooming, and many factors geometry cannot capture. High-scoring celebrities typically score 7–8, not 10.
Why does my score change between photos?
The tool calculates from 468 facial landmarks detected in a single photo. Lighting, angle, facial expression, and head tilt all shift landmark positions, which directly affects measured distances and ratios. A front-facing, neutral-expression photo in good light gives the most consistent result.
Is my photo stored or sent anywhere?
No. The entire analysis — face detection, landmark mapping, and scoring — runs locally in your browser using WebAssembly. Your photo is never sent to any server and is cleared from memory when you close or refresh the page.
What AI model does Pretty Scale use?
Pretty Scale uses MediaPipe FaceLandmarker, which detects 468 facial landmarks from the photo using a neural network that runs in WebAssembly. The scoring algorithm applies geometric calculations to those landmarks entirely client-side.
Why am I getting an error that no face was detected?
Face detection fails when the face is too small, too blurry, heavily shadowed, at an extreme angle, or partially obscured. Try a larger, clearer, front-facing photo with even lighting on both sides of the face.