Asymmetric Face: Causes and What Helps
Facial asymmetry is normal. See what causes it, which habits evidence does not support, and how to measure your left-right difference from a photo.
Read guideGet a free 1–10 AI face rating with independent facial measurements. Your photo is analyzed in your browser and never uploaded.

Generated sample. AI face score: 8.7/10. Face shape: Oval. Symmetry: 99/100. Proportions: 0.83:1. Jawline: 77%.
A face rating is a numerical estimate of appearance from a photo. Different tools use different models, crop rules and scales. Ours gives a local image-model estimate out of 10 alongside independent geometric observations, so you can examine the photo from more than one angle.
Ours pairs a local image-model estimate with independent measurements of symmetry, proportions, shape and eye angles. When usable landmark analysis is available, explore them together without treating the measurements as a breakdown of the model’s score.
It is a photo-based estimate from one image. Expression, lighting and framing can affect it; personality, chemistry and your look in motion are outside this snapshot. Use the details as visual reference for style and proportion decisions.
Read it as one photo-based reference: repeatable under the same conditions, but only one part of the overall picture.
Review your face rating from one clear photo, save a local result card, or explore another free tool for a closer look at your measurements.
Most face rating tools use a 1–10 scale, and ours works the same way. Here’s how to read the number without turning it into a ranking.
| Score | Position on the scale | What it reflects |
|---|---|---|
| 1.0–3.9 | Lower values | Lower output from the image model; not a diagnosis or a description of a specific feature |
| 4.0–6.9 | Middle values | The central part of the numerical scale, not a measured population average |
| 7.0–9.9 | Higher values | Higher output from the image model; not a percentile or a promise of how others see you |
| 10.0 | Upper endpoint | The maximum display value, not a claim of perfect proportions |
The midpoint is a position on this scale, not a population average. Scores from different tools are not interchangeable: models, photo processing and scales differ. No number here tells you your percentile or what another person will think of you.

Get a 1–10 AI estimate from your photo, with separate readings for symmetry and facial proportions. The analysis runs locally in your browser.
A 478-point mesh levels the eye line and frames a consistent crop.
An on-device image model scores the aligned crop on a 1–10 display scale.
Symmetry, thirds, fifths, and contours are measured independently.
First, an on-device MediaPipe model detects a complete facial landmark mesh. Four landmarks around each eye help level the eye line and correct sideways roll. The pipeline creates a 256-pixel similarity crop using the forehead-to-chin span as its scale reference, then takes the fixed 224 × 224 model input without stretching the features.
A local MobileNetV2 image model evaluates that crop and its horizontally mirrored copy. We average the two photo-based predictions, map the result to the 1–10 display scale, clamp it to that display range and round to one decimal place. There is no geometric weighting, bonus point or post-hoc score floor. The model was trained on human photo ratings, so the output reflects learned preferences rather than a universal beauty standard.
The AI model runs locally in your browser. Your photo is never uploaded for the free score.
Separately, the report computes four primary geometric readings: left-right balance, facial thirds, facial fifths and face proportion (cheek width divided by forehead-to-chin height). When usable landmark analysis is available, the same landmark logic can also show face shape, eye-corner angle, symmetry groups and lower-face outline. These are visible geometric observations, not weights or explanations for the AI score.
The same image under the same processing conditions gives a repeatable result. Change the lighting, expression, camera distance or head position and the reading can change. A front photo also cannot measure chin or nose projection reliably; those need a side view.
The score is the local image model’s estimate for the visual patterns in this photo. The geometric readings shown beside it are separate observations and do not explain points added or lost.
Three things it is not:
The bands themselves live in one place on this page: the face rating scale.
The score is measured from the photo, not from you. Keep the capture conditions steady so the model and landmark readings have a more consistent image to work with.
Even a slight head tilt can shift left–right measurements.
A close camera can enlarge the nose and squeeze the edges of the face.
Face a window so both sides of your face receive similar light.
Keep your eyes and face outline clear so the browser can locate them.
Keep the same setup when comparing photos. Choose a readable JPG, PNG, or WebP under 10 MB, up to 40 megapixels, with one clear face. An analysis is withheld if the file cannot be decoded, no face or multiple faces are detected, the image is under 480 × 480 pixels, lighting or contrast fails quality checks, a complete facial landmark mesh cannot be detected, or model inference fails.
“Accurate” only means something once you say accurate about what. Here, the score is an AI estimate from one image, while the geometric measurements describe detected landmarks. Neither establishes an objective attractiveness score or a measurement of how you look in motion, in conversation or to any one person.
Read the result as a photo-specific reference. The model reads the pixels available in the selected image; the separate geometry describes visible landmark relationships. Neither tells you what other people think of you.
Do not compare scores across websites. Different models, crop rules, scales and photos can produce different numbers. If a number starts shaping how you feel about your face, take a break rather than chasing a different result.
Review the photo tips when you want to make a more consistent comparison.
Put the perspective to practical use.
Facial asymmetry is normal. See what causes it, which habits evidence does not support, and how to measure your left-right difference from a photo.
Read guideSee how camera distance, lighting, angles, expression and filters change the way you look in photos.
Read guideExplore 10 free tools for facial proportions, eyes, jawline, profile and photo age.
PSL-style bands from 4 weighted measures.
Inner-to-outer eye corner angle in degrees.
Compare phi (1.618), thirds and fifths.
Eye opening ratio, corner tilt and spacing.
Closest shape match from 478 landmarks.
Frontal jaw angle, taper and contour edge.
Left-right balance across 28 landmark pairs.
Side-profile jaw corner angle in degrees.
On-device perceived age range from a photo.
Side-profile chin projection and alignment.
The test, the score, your photo and the limits worth understanding.
The browser aligns your photo using four landmarks around each eye, creates a 224 × 224 model input after the alignment crop, and runs a local MobileNetV2 image model on the crop and its mirrored copy. The two photo-based predictions are averaged and mapped to the displayed 1–10 scale, rounded to one decimal place. Separately, when usable landmark analysis is available, deterministic geometry measures proportions, symmetry, face shape, eye angles and lower-face ratios. These readings are independent observations, not components of the 1–10 score.
No. Your photo, landmarks, score and downloadable PNG result card are processed locally in your browser using JavaScript and WebAssembly. We never upload or store your image or results. Our analytics record only aggregate, non-personal events such as page type and whether a scan succeeded or failed.
Different websites use different models, crop boundaries and scoring scales, so numbers across tools are not comparable. Even on this page, changing camera distance, lighting angle, head pitch or expression changes the pixels and landmarks available to the model, so the result may move even when your face has not changed.
We do not publish a population average or percentile ranking. Because results are not collected from users’ devices, there is no central database of scores. A 5.5 is simply the mathematical midpoint of the 1–10 display scale.
The analyzer accepts a readable JPG, PNG, or WebP under 10 MB, up to 40 megapixels, showing one clear face where a complete facial landmark mesh can be detected. A photo is rejected if it fails resolution, lighting, or contrast checks, or if no face, multiple faces, or an incomplete landmark mesh appear.
No. Because we keep no server copy, closing or reloading the page clears the workspace. Before leaving, use Save result card, which exports a local PNG without your photo by default, or Copy summary to keep your readings.
No. PSL is a separate community rating system with different conventions. Our output is a local image-model estimate on a 1–10 display scale, and numbers across these systems are not interchangeable. Explore a separate geometric breakdown with our PSL Score Test.
Face ratings sometimes appear in looksmaxxing discussions as before-and-after numbers. This site provides local, style-oriented photo references for exploring hair, glasses and framing—not a reason to change your face or pursue a medical procedure.
Four papers inform how we describe proportions, preferences and photo distortion. They provide context; they do not validate our scoring formula.