Face Symmetry Test
Measure left–right facial balance from one straight-on photo, using a 478-point landmark detector — entirely in your browser.
or drag one here
Your photo is analysed on this device and never uploaded.
How the test works
Everything happens on the device in front of you. The photo you upload or capture is handed to a face-landmark model that runs inside the browser tab, and it returns 478 points describing the geometry of the face in that single image — the outline of each eye, the ridge of each brow, the bridge and wings of the nose, the edges of the lips, the line of the jaw. Nothing about that photo travels anywhere else; the measurement that follows is arithmetic on those 478 points, done locally.
The first step is finding the face's own center line, not assuming the photo is perfectly framed. Five points near the vertical center of the face — the upper forehead, the point between the brows, the bridge of the nose, the base of the nose, and the bottom of the chin — are fit to a single straight line using orthogonal regression, which minimizes the perpendicular distance from each point to the line rather than the vertical distance. That line becomes this face's fitted axis, and every later measurement in this photo is made against it instead of the image's own center or a raw guess at "straight up."
Before anything is scored, the same detection is used to estimate how far the head is turned away from facing the camera. A photograph of a turned head measures further apart from itself even on a face with no underlying asymmetry at all — turning introduces its own geometry. Past roughly three degrees of estimated turn, this tool withholds the overall score rather than publish a number that pose, not anatomy, would be driving.
Within that gate, twelve landmark pairs across five regions — eyes, brows, nose, mouth and jaw — are compared. Each pair's first point is reflected across the fitted axis, and the distance from that reflected position to where its partner point actually sits is the residual for that pair. Every residual is then divided by the distance between the inner corners of the two eyes, so the result does not depend on how large the face is in the frame or how close the camera was. Each region's residuals become a score, and the five region scores combine into one overall number, weighted toward the regions this detector locates most reliably.
What facial symmetry is
Facial symmetry, in the sense this tool measures it, is how closely the left and right sides of a face mirror each other across a line down its center — the fitted axis described above, not necessarily the exact vertical center of a photograph. It is a purely geometric comparison: where a landmark on one side sits, reflected, against where the equivalent landmark on the other side actually is.
No human face is a perfect mirror image of itself, and that is not a defect to be corrected. Skulls grow slightly unevenly from childhood on, chewing and expression favor one side over years, old injuries and everyday habits — which side you sleep on, which side you chew on — all leave small, permanent traces. Even identical twins, who share the same genome, do not have identical faces. What varies between people, and between photos of the same person, is only how large those ordinary differences are and how visible they happen to be in a given image.
Because of that, "symmetric" and "asymmetric" are best read as ends of a continuum that every face sits somewhere along, rather than two categories a face belongs to. This tool reports where a single photo landed on that continuum, for twelve specific pairs of points — nothing more sweeping than that.
It is also worth being clear about what this measurement is not. It says nothing about which side of a face is the "correct" one, because there isn't one — a fitted axis is simply the line the two sides happen to sit closest to in this photograph, not a blueprint either side deviated from. And it is a measurement of position, not of health, muscle strength or function; a face that moves unevenly when it speaks or smiles can still show a high score at rest, and a face with a high resting score can still move unevenly, because those are different things being measured in different ways.
What your score means
The overall number is a weighted combination of five region scores, each built the same way: the closer the reflected landmarks land to their real partners, relative to the distance between the eyes, the higher the region scores. That combined number is then placed into one of four bands, which exist only to describe this tool's own output in plain language — they are not percentiles, they are not drawn from any population of other people's photos, and they are not a clinical scale of any kind.
| Score | Band |
|---|---|
| 90–100 | Very close match |
| 80–89 | Close match |
| 70–79 | Moderate match |
| Below 70 | Larger measured difference |
Treat the number as a description of one photo, not a verdict on a face. The most useful comparison it supports is with another score of yours, taken under matched conditions — straight to the camera, neutral expression, even light — because that is the comparison where the number is measuring the same thing twice. Comparing it against someone else's score, or against a remembered "good" number from a different tool, compares two things that were never built to line up: different faces, different cameras, different lighting, and in this tool's case, bands invented for its own scale rather than any shared standard.
A score can also change between two photos of the same face taken minutes apart, and that swing usually says more about the two photographs than about the face — a slightly turned head, a different expression, a lens held a little closer. None of that is the tool malfunctioning; it is the tool correctly measuring two different pictures.
Underneath the bands, the raw number does not move in a straight line as a measured gap grows. It responds most sharply near a perfect match: a very small gap between a reflected landmark and its real partner already costs several points, because the scoring curve is steepest there. Further down the range, the same size of additional gap moves the number by less — the curve flattens out, rather than continuing to fall at a constant rate. Two photos with visibly different amounts of measured difference can therefore land in the same band, and two photos that look similar at a glance can land in different ones. The four bands exist to make a continuous number easier to describe in a sentence; the number underneath them is what actually changes between photos.
The five feature scores
The overall score is built from five region scores, and each is worth expanding on individually — what it actually compares, and what ordinarily moves it, without claiming to know which of those causes is behind any particular result.
Eyes
Four points around each eye are compared: the outer corner, the inner corner, the highest point of the upper lid, and the lowest point of the lower lid. Because the eyes sit close to the fitted axis and their edges are high-contrast against skin, this is the region the underlying detector locates most consistently — which is why it carries the largest share of the overall weighting. What commonly moves this score: a slight head tilt or turn, one eye opened wider than the other in a candid expression, or asymmetric lighting that changes where a lid's edge appears to fall.
Brows
Three points per brow are compared: where it begins near the nose, its highest arch, and where it tails off toward the temple. Brows move more than almost any other landmark on the face — they are the primary muscle used to signal surprise, concentration, or skepticism — so this score is unusually sensitive to the exact instant a photo was taken. A raised brow, a slight frown, or brows simply grown in unevenly can all shift this number independent of anything else on the face.
Nose
A single pair of points at the outer edges of the nostrils is compared here, which is why this region carries the least data of the five and is treated as one supporting signal rather than a detailed map of the nose. Genuine, permanent structural differences between nostrils are common and are exactly what this pair is built to catch, but flash photography, close framing and a slightly turned head all shift where these edges appear to sit in a single image.
Mouth
The two corners of the lips are compared. This is a resting-face measurement, not a smile: an unevenly deployed smile and a resting mouth engage different muscles and do not necessarily move together, which is why this tool asks for a neutral expression rather than scoring a smiling one. What commonly moves this score: any expression at all, the natural resting position of the mouth, and how the head is tilted at the moment the shutter opens.
Jaw
Three points along each side are compared: the temple, the cheekbone, and the point where the jaw's outline curves toward the chin. This is the widest-tolerance region of the five and, for a related reason, the one weighted second: its outline is the hardest of the five for a detector to place precisely, because hair, shadow and the angle of the jaw itself all shift where the detected edge falls between one photo of a face and the next, even with nothing about the face having changed.
Getting an accurate result
A landmark detector can only measure the photo it is given, so the conditions a photo is taken under matter more here than in an ordinary snapshot. The following six habits remove most of the distortion this kind of measurement is vulnerable to, roughly in order of how much difference each one makes.
- Face the camera straight on. A turned head is the single biggest source of a misleading result — it is also the one condition this tool checks for directly and will withhold a score over, because a turned head changes the measured geometry regardless of the face underneath it.
- Keep a neutral, relaxed expression. A smile, a raised brow, or a squint activates the two sides of a face unevenly by design, and this tool is built to measure the resting structure underneath those movements, not the movements themselves.
- Hold the camera at eye level. Shooting up from below or down from above foreshortens one part of the face relative to another, in a way that a straight-on shot from eye height does not.
- Light the face evenly from the front. Light from one side casts a shadow across the other, and shadow can move where a detector places an edge — an effect of the light, not the face beneath it.
- Pull hair back off the forehead and jaw. Hair covering the brow line or the edge of the jaw hides exactly the outline this tool is trying to locate.
- Remove glasses before the photo. Frames and lenses sit directly over the eye landmarks this tool depends on most heavily, and reflections in the lenses can confuse the detector further.
None of this changes anything about a face — it only removes conditions that would have changed the measurement of it. A well-lit, straight-on, neutral photo is simply the version of a photo this kind of measurement was designed to read.
The mirror test
The mirror test is an old idea, usually done by hand: take a straight-on photo, cut it down the center, and pair each half with its own reflection to build two whole faces — one made entirely from doubled left halves, the other entirely from doubled right halves. Because each composite repeats one side of the face, small differences between the two original sides become much easier to see once they appear twice in the same image.
This tool builds both composites the same way, with one change to how the center line is chosen. Rather than cutting down the middle of the photograph — which assumes the photo itself is perfectly centered and perfectly level — it cuts along the fitted axis described earlier, the line this tool locates from five points on the face itself. A photo that is a little off-center or a little tilted still produces two composites cut along the face's own center, not the image's.
Neither composite is a hidden "real" face waiting to be uncovered, and neither is more accurate than the original photo — both are built by doubling one half, which is itself a form of exaggeration. People commonly report preferring how they look in the original, unaltered photo to either mirrored version, when shown all three side by side. The value of the comparison is in the gap between the two composites: a small gap suggests the two sides of that photo were already close together; a larger gap suggests more measurable difference between them, for reasons this visual alone cannot separate out.
The two composites are also more sensitive to a turned head than the overall score is, because doubling one side doubles whatever a turned head does to that side too. A composite pair built from a photo taken a little off-axis can look more dramatic than the same face's actual measured result would suggest, which is part of why this tool treats the mirrors as a visual companion to the score rather than a second score in their own right — there is no band, no number and no pass or fail attached to either composite, only the image itself.
What this test cannot tell you
A single two-dimensional photograph cannot, on its own, separate a face's underlying structure from everything else that shaped that particular image: how far the head was turned, how close the lens was and what focal length it used, where the light was coming from, and what the face happened to be doing at the instant the shutter opened. All four of those move the measured position of a landmark exactly as a real anatomical difference would, and no arithmetic performed after the fact can fully untangle them from a single frame. That is the reason for the pose gate described above, and the reason repeat photos of the same face can land in different bands — the tool is reading the photograph in front of it, faithfully, and a photograph is not the same thing as the face.
It follows that this tool does not, and cannot, assess health, function or appearance — only the geometric position of twelve point pairs in one image. It is not a medical device, was not built or tested as one, and a score from it is not evidence about any underlying condition. If an asymmetry is new, is getting more noticeable over time, is painful, or comes with any change in how part of the face moves or feels, that pattern is worth describing to a qualified clinician, in person — not something a browser tool can rule in or out from a photo.
Frequently asked questions
What is a good face symmetry score?
This tool sorts scores into four of its own bands: 90 and above is a very close match, 80 to 89 a close match, 70 to 79 a moderate match, and below 70 a larger measured difference. These are labels this tool invented for its own output, not a clinical scale and not a comparison against anyone else's photo — there is no population of other results behind them. A score only means something next to another score of yours taken the same way: straight on, neutral expression, even light. Comparing it to a remembered "good" figure from somewhere else is not what these bands are for.
Is anyone’s face perfectly symmetrical?
No. Every face carries some left-right difference — in bone growth, muscle use, old injuries, even which side you sleep on — and that is ordinary, not a flaw to correct. A score near the top of this tool's range does not mean the twelve measured landmarks were identical; it means the differences between them, expressed as a fraction of the distance between your eyes, were small enough in that one photo to fall in the highest band. A different photo of the same face, taken minutes apart, can land in a different band without anything about the face having changed.
Why does my face look more asymmetric in photos than in the mirror?
A mirror shows you your face flipped, and it is the version you have seen thousands of times — brushing your teeth, getting ready, glancing at a window. Psychologists call the resulting preference the mere-exposure effect: familiarity itself reads as more pleasant. A photograph or a video call shows the unflipped orientation, so whatever asymmetry you already have now sits on what feels like the "wrong" side. Nothing about your face changed between the two views. This is also why the mirrored composite this tool produces can look more familiar to you than the un-mirrored one, or than the original photo.
Does facial symmetry equal attractiveness?
No — research on face perception counts symmetry as one input among several, alongside averageness, proportion, skin texture, expression and plain familiarity, and its measured effect on attractiveness ratings is real but small and levels off well before perfect symmetry. Two of the most symmetric faces on paper can be judged very differently once expression, lighting and context are added, and a face rated highly attractive is very rarely perfectly symmetric to begin with. A number from this tool describes a geometric measurement of one photograph. It is not a verdict on how anyone looks, and it was not built to produce one.
Can I improve my facial symmetry?
It depends entirely on what is causing the difference, and a single photo score cannot tell you that — it reports a result, not a diagnosis. A widely cited 2018 study of facial-exercise routines measured changes in cheek fullness and perceived age after several weeks of exercises; it did not measure symmetry, and its results should not be read as evidence that exercises change symmetry. The bone structure an adult already has does not change with exercise. If an asymmetry is new, getting more pronounced, painful, or affecting how your face moves, that is worth discussing with a qualified clinician — not something a browser tool can assess.
Why is my smile crooked in photos?
The two sides of a smile are driven by separate muscles and rarely fire in exact lockstep, so most smiles are a little uneven — that alone is not a problem. Add a slightly turned head, a camera not quite at eye level, timing the shutter mid-motion, and ordinary bite differences, and a crooked-looking smile is easy to produce on an otherwise balanced face. This tool deliberately scores a neutral, resting expression rather than a smile, because expression moves the measured landmarks independently of the underlying structure. A sudden new weakness on one side of the face, especially with other symptoms, needs prompt medical attention, not a photo test.
How accurate is this face symmetry test?
This tool has not been validated against a clinical measurement standard, so no accuracy percentage is published for it, and none should be trusted from a browser tool that hasn't been. What it does is described plainly: it detects 478 facial landmarks, scores twelve bilateral pairs against an axis fitted from five midline points, and withholds the overall score whenever the estimated head turn exceeds three degrees, because a turned head produces geometry that looks like asymmetry but isn't. Even within that gate, expression, hair covering the jaw, uneven lighting and lens distance at close range can all move the result on an otherwise valid-looking photo.
Is my face asymmetrical, or is it just the camera?
Often, partly the camera. A phone's front camera is a wide-angle lens used at very close range, and a wide lens exaggerates whatever is nearest it — turn your head a little and one cheek or one side of the jaw balloons relative to the other, an effect of the optics, not the face. The mirrored preview most phones show while you frame the shot flips the image again, which can make a real but minor difference look like it's on the opposite side. Shooting with the rear camera from arm's length, held at eye level, removes most of this distortion before any measurement begins.
What is the face symmetry mirror test?
The classic version reflects one half of a face across the center of the photo to build two whole faces, each made of two identical halves, so you can compare them side by side. This tool builds both composites from the fitted landmark axis — the line through five midline points this tool locates on your actual anatomy — rather than from the geometric center of the image, so a photo that isn't perfectly centered or perfectly level doesn't skew the result on its own. Neither composite is a hidden "true" face; both are exaggerations, built to make existing left-right differences easier to see, not to reveal one.
Why do my two mirrored faces look like different people?
Because mirroring doubles whatever a side already carries. If your left eye sits a touch higher than your right, the composite built from doubled left halves shows two eyes both a touch higher, and the one built from doubled right halves shows the opposite — small, ordinary differences become obvious once repeated across the whole face. The wider the gap between your two composites looks, the more left-right variation was present in that photo, which is a mix of real anatomy, head turn and lens distance, not a hidden feature being uncovered. People commonly report preferring how they look in the original, unmirrored photo to either composite.