Facial harmony, symmetry, and proportion analysis, in under a minute
LooksLab is a face analysis tool for people who want real numbers instead of vibes. Two photos in, you get scores for Harmony, Angularity, Dimorphism, and Health Indicators, plus a breakdown of how each feature compares to a population baseline. Landmark detection runs in your browser. Scoring runs on our server using only the landmark coordinates, your photo is not sent to the scoring endpoint.
How it works
The pipeline has three stages. None of them ask for measurements or for you to mark points by hand.
1. Upload a front and a side photo
The front photo should be straight on with a relaxed expression and your hair off your forehead. The side photo should be a true profile, the same way you would shoot a passport photo if you rotated ninety degrees. The upload step auto-crops both shots and checks that the face is centered and large enough for the landmark detector to lock on. If something is off, you get a specific reason rather than a generic error.
2. Computer vision detects about 478 landmarks
We use Google's MediaPipe Face Mesh to find around 478 points across the jaw, cheek, brow, nose, eyes, and lips on each photo. The model runs in WebAssembly inside your browser, so the photos never leave the device for this step. The output is a dense mesh that captures shape rather than appearance, which is what makes the rest of the scoring possible.
3. Scoring against research-backed curves
The scoring engine extracts 66 ratios and angles from the mesh, things like canthal tilt, midface to lower face ratio, jaw angle, and ogee curve. Each one is mapped through a cubic-bezier curve calibrated against population data, so a measurement near the population mean scores high and tails fall off smoothly. This step runs on our server, which only receives the landmark coordinates and your selected gender, not the photo itself. The individual scores combine into the four pillar totals (Harmony, Angularity, Dimorphism, Health Indicators) and also feed the per-metric breakdown.
What you get
The result page is organized around four pillars. You can expand any of them to see the underlying measurements and how each one was scored.
Harmony
How close key proportions and ratios are to population-calibrated targets. Includes vertical thirds, horizontal fifths, midface ratio, lower-third proportion, eye separation, and similar geometric measures.
Angularity
Bone-driven structural sharpness. Jaw angle, chin projection, cheekbone definition, and brow ridge geometry. Angularity is independent of harmony, so you see a separate read on whether your face leans more sculpted or more rounded.
Dimorphism
How strongly features sit on the typical masculine or feminine end of population distributions for your selected gender. Higher scores mean features land closer to the modal range for that gender across the calibration data.
Health Indicators
A small set of measurements that have been associated with general health and developmental signals in the literature. These are correlates, not diagnostics, and should be read that way.
Privacy first
Landmark detection runs entirely in your browser through MediaPipe and WebAssembly. The scoring step that follows runs on our server, but it only receives the landmark coordinates and your selected gender. Your photo is not sent to the scoring endpoint.
Photos are uploaded to a private Supabase Storage bucket so you can revisit your scan and we can render the annotated front and side views. This happens for both anonymous and signed-in sessions. Anonymous scans are scoped to a session-bound identity. Signed-in scans are scoped to your Google account. Either way, your scans show up on the My Scans page, and each one has a delete action that removes the database row and the underlying images.
We do not sell your data, we do not train models on your photos, and we do not share scans with third parties. Analytics events are aggregate and do not include the photo data. If you want a full account wipe, you can request it from the account page and the delete cascades across every scan, photo, and profile row.
Methodology
The scoring engine uses 66 calibrated curves, each one corresponding to a specific ratio, angle, or distance derived from the landmark mesh. Every curve is a cubic-bezier defined by custom control points, calibrated so that the population mean for that measurement sits near the top of the curve and the tails fall off at a rate that matches the observed spread. The four pillar scores are weighted blends of the underlying curves rather than a single number.
The same pair of photos will give you very close to the same scores on every run. The variance you do see usually comes from head tilt, camera distance, or lens choice, which is why the upload step enforces a reasonable framing before the landmark step can succeed.
For a deeper walk through the curves, the calibration data, and the weighting choices, see the methodology page.
Frequently asked questions
How does pricing work?
You can try the scan without paying. Anyone gets a free preview of their top strengths and weaknesses. The four pillar scores, the overall score, and the full per-metric breakdown unlock with a subscription.
How accurate is the analysis?
Landmark detection uses Google's MediaPipe Face Mesh, which places roughly 478 points on the face. Scoring runs those points through curves calibrated against a population dataset, so results are repeatable: the same pair of photos will give you very close to the same numbers each time. Lighting, head tilt, and lens distortion can shift individual measurements, which is why we ask for a clean front photo and a true profile.
Do you keep my photos?
Photos are uploaded to a private Supabase Storage bucket as part of the scan flow, for both anonymous and signed-in sessions, so you can revisit your scan and we can render the annotated views. Each scan is scoped to your account (Google sign-in) or to a session-bound anonymous identity. You can delete any scan from the My Scans page, which removes both the database row and the stored photos. We do not train models on your photos.
Can I use it on a phone?
Yes. The scan flow is designed mobile-first. The landmark detection runs in your browser, so a recent iPhone or Android with a modern browser is enough. No app install is required.
What's the difference between harmony and angularity?
Harmony measures how close a given proportion is to a target value drawn from population data, for example the ratio of midface to lower face height. Angularity measures the sharpness of features like the jaw, cheekbone, and brow ridge. A face can score high on harmony but low on angularity, or the other way around, and the breakdown shows both side by side.
Is the golden ratio actually predictive?
The golden ratio is one input among many. It correlates with perceived attractiveness in some studies but explains only a small share of the variance on its own. LooksLab uses 66 separate calibrated curves covering thirds, fifths, canthal tilt, ogee curve, jaw angle, and many other features. The overall score blends all of them rather than collapsing your face to a single ratio.
Do I need to sign in?
No. You can complete a scan as a guest. Signing in with Google saves your scan history, lets you compare scans over time, and syncs your calibration settings across devices.
Can I delete my data?
Yes. Each saved scan has a delete action that removes the database row and the underlying photos. You can also remove your entire account from the account page, which cascades and deletes every scan, photo, and profile row tied to your user.