How we measure this

We use MediaPipe FaceLandmarker, Google's open-source, peer-reviewed computer-vision model. It detects 468 facial landmark points on your photo directly in your browser via WebAssembly. Your image is never uploaded to a server — the entire detection runs on your device.

The following metrics are computed directly from the detected landmark coordinates on your photo and are tagged measured in your report:

• · Jawline definition — mandibular contour angles derived from jaw landmarks.
• · Cheekbone prominence — zygomatic arch geometry from midface landmarks.
• · Facial symmetry — bilateral landmark alignment relative to the vertical midline.
• · Canthal tilt — angle between the medial and lateral canthus of each eye.
• · Facial thirds — hairline-to-brow, brow-to-nose-base, nose-base-to-chin heights as percentages of total face height, and deviation from a balanced 33/33/33.
• · FWHR (facial width-to-height ratio) — bizygomatic width divided by upper-face height (brow to upper lip).
• · Gonial angle — the actual mandibular corner angle in degrees, averaged left and right.
• · Jaw:cheek width ratio — bigonial ÷ bizygomatic width (how V-shaped vs square the lower face reads).
• · Interpupillary-distance ratio — eye-center spacing divided by cheek width. We use a medial+lateral canthus midpoint as an eye-center proxy (no dependency on iris landmarks).
• · Mouth-to-nose width, philtrum ratio, lip fullness — front-view ratios from mouth, alar-base, subnasale and lip vermilion landmarks.
• · Hunter-eyes composite (0–100) — weighted combination of canthal tilt, brow-to-eye distance, and upper-lid exposure.
• · Jaw-definition composite (0–100) — weighted combination of gonial angle, jaw:cheek ratio, and jawline score.
• · Face shape — best-effort oval/round/square/heart/oblong/diamond classification from width and height ratios. Heuristic, not a diagnosis.
• · Golden ratio exploration — the raw distance of a few proportion pairs (face height:width, mouth:nose, mid-face:lower-face) from φ ≈ 1.618. Shown as an optional educational section only: the golden ratio is a historical aesthetic concept, not a validated measure of attractiveness, and it does not feed any HarmonyFace score, tone, or recommendation.
• · Nasofrontal & nasolabial angles, chin projection — computed from a quarter-profile photo when provided.
• · True side profile (optional) — if you upload a full 90° lateral shot and detection succeeds, it's preferred over quarter-profile photos for E-line, mentolabial, nasofrontal & nasolabial angles, and chin projection, since a true lateral view avoids the perspective distortion inherent to 3/4 angles. MediaPipe's model is trained primarily on frontal/near-frontal faces and is not expected to reliably detect faces at a full 90° angle — failures are a known limitation of the underlying model, not a bug in this app. Quarter profiles remain the reliable fallback when the side photo can't be read.

Some parts of the report use reference data rather than measurements from your photo. We flag these as illustrative and don't tag them as measured:

• · Skin clarity, texture, and blemish scoring — requires a dedicated skin-analysis model we haven't shipped yet.
• · Treatment priority weighting — currently uses rule-based reference data pending a validated clinical model.
• · Mentolabial angle and E-line — measured from a quarter-profile photo when provided; clinically calibrated versions still require standardized quarter-profile photography.
• · Mandibular plane angle and nasal tip projection — deliberately not shipped. Both require a stable ear-tragus or Frankfort-horizontal reference that our quarter-profile landmark set doesn't provide accurately enough to be trustworthy.
• · Hair & hairline, smile, ears, neck — no landmark model shipped for these; we show general guidance and (where you uploaded them) your Front Smiling / Top View photos as references.
• · Dimorphism and prototypicality scoring — intentionally omitted. These require population reference datasets we don't have, and can't be honestly faked client-side.

Price ranges shown for treatments (Morpheus8, Sculptra, Botox, fillers, etc.) are examples based on published US clinic pricing, not live quotes. Actual pricing varies by geography, provider, and case complexity. Always get a written quote from a licensed provider before booking.

If you fill in the optional "About you" fields on upload (population group and sex), four profile metrics — nasofrontal angle, nasolabial angle, labiomental (mentolabial) angle, and mouth-to-nose width ratio — are compared against per-group posterior means from Wen et al. 2015, Inter-Ethnic/Racial Facial Variations: A Systematic Review and Bayesian Meta-Analysis of Photogrammetric Studies, PLOS ONE (Tables 2 & 3). Only three broad research groups (African, Asian, Caucasian) and a male/female split are covered, because those are the only groupings that paper publishes usable numbers for. Every other metric in your report keeps its existing universal reference range; if you skip either field, all metrics fall back to the universal average.

Photos are processed entirely in your browser. Nothing is uploaded, stored on a server, or shared. Saved reports live in your browser's local storage on your device.

HarmonyFace is an educational tool. It is not a diagnostic device and does not replace a consultation with a board-certified dermatologist, plastic surgeon, or cosmetic physician. Landmark detection accuracy depends on photo quality, lighting, angle, and facial expression. Nothing in this report should be interpreted as a medical recommendation.