Face Scan Scores Change With Age: What Actually Moves

The short answer

Your face at 45 is not your face at 25 with different numbers attached. The bone underneath is mostly fixed by the mid-twenties. Everything on top of it keeps changing, slowly and in one direction. A face scan measures distances between points on that surface, divides them into ratios, and maps the ratios onto a scale. When the surface moves, the ratios move. The tool reports the movement honestly and has no idea what caused it.

That is the whole problem in one sentence: a score is a measurement of geometry, and geometry is not stable across a lifetime.

What actually changes, and when

It helps to separate the layers, because they age at very different rates.

Bone. Growth of the facial skeleton stops in the late teens to early twenties. After that, jaw width, cheekbone width and the distance between your pupils are essentially constants. A scan that reports a change in these is reporting a change in detection, not in your skull.

Fat. The midface fat pads lose volume and slide downward. Temples hollow, the area under the cheekbone flattens, and the region under the eye becomes less full. This is the single largest visible change between 35 and 55, and it is soft tissue, not bone.

Skin. Collagen density declines gradually from the mid-twenties onward. The practical effect for a scan is that edges get softer: the line where the lip meets the skin, the crease of the eyelid, the shadow along the jaw. Landmark models find edges, so softer edges mean less precise points.

Eyelids. Hooding reduces the visible opening of the eye. Eyelid landmarks are already the least stable points in most models, because the eye contour changes with blink, expression and light. Add age and the error grows.

Lips. The vermilion thins with age, so lip height and lip-to-chin ratios fall even when nothing else has changed.

Cartilage. Nose and ear cartilage continues to change slowly through adulthood. The tip of the nose can broaden slightly over decades. It is a small effect, but it sits in the middle of the face where several ratios are anchored.

Why the score moves more than the face does

Two mechanisms amplify small changes.

The first is division. Almost every face measurement is a ratio: a distance divided by another distance. If the numerator drops ten percent because an eyelid is lower, the ratio drops ten percent. But the mapping from ratio to score is usually linear between two endpoints that were chosen by the developer, and that line is steep in the middle. Ten percent of geometry can become fifteen points of score.

The second is the reference sample. A formula that turns a ratio into "72 out of 100" is calibrated against some set of faces. If that set skews young, an older face sits far from the middle of the distribution, and a percentile-style output punishes the distance twice: once for the actual difference, once for being outside the range the developer tuned for.

Neither mechanism is a bug. Both mean the same thing: the number is relative to a reference you cannot see.

Which results move first

The general rule is that results built on soft-tissue edges move before results built on bone landmarks.

• Eye-area measurements move first, because hooding and under-eye hollowing change the contour the model traces.
• Mouth and nasolabial measurements move next, because lip volume and fold depth both change.
• Jaw contour moves after that: the bone corner is stable, but the soft tissue over it is not, and many models place the jaw point on the soft edge.
• Cheekbone width and interocular distance barely move at all, because they are bone-to-bone spans.

Symmetry is a special case. Facial bone asymmetry does not increase much with age, but soft-tissue asymmetry does: one eyelid droops earlier, one side of the mouth loses volume faster. A symmetry score computed on soft-tissue points will therefore drift downward even though the skeleton is unchanged.

What the number cannot tell apart

Four different causes produce exactly the same output on the screen:

1. Real aging.
2. Weight change of a few kilograms, which alters cheek fullness and jaw definition.
3. Dental work, which changes lower-face height and lip support.
4. Setup: overhead light deepens every fold, a lower lens angle shortens the jaw, a wider lens widens the midface.

A scan sees a photograph. It has no way to separate these. So "my score dropped six points" does not mean "I aged five years". It means one of those four things happened, and three of them are reversible within a week.

How to compare two scans fairly

If you want to track change over years rather than minutes, the protocol matters more than the tool.

1. Fix the camera. Same phone, same distance (about 60 cm), lens at eye height, rear camera if you can, no zoom.
2. Fix the light. One window, daylight, facing it, no overhead lamp, no direct sun.
3. Fix the face. Hair off the forehead, glasses off, neutral expression, no makeup if you wear it inconsistently.
4. Take three, keep the median. A single scan is a sample, not a measurement.
5. Space them out. Quarterly is enough. Weekly retesting measures your sleep, not your face.
6. Log the confounders. Weight, dental work, a new pair of glasses, a course of steroids, a bad week of sleep.

Then read the trend, not the pairs. A line that drifts two points over a year is noise. A line that drops eight points over six months, with weight and setup unchanged, is worth a second look — and worth taking to a mirror rather than to the app.

Three mistakes that make the drift look worse

Comparing to an old selfie from a different phone. Front cameras changed focal length between 2018 and 2024. A photo taken at 26 mm and one at 24 mm are not the same face. The selfie angle guide covers what the lens does to a face before any scan runs.

Retesting until the number improves. If you scan five times and keep the best, you have measured your persistence, not your face. The variance between retakes is usually larger than a year of real change.

Using a younger photo as the baseline. That is not a baseline, it is a different face. If you want a reference, take a fresh set today and compare forward.

What is still worth measuring

None of this makes the numbers useless. It makes them local. A scan is decent at telling you whether two photos of the same person, taken the same way, differ. It is bad at telling you what a face is worth, and worse at telling you what it will be worth in ten years.

For the parts that barely change — face shape, proportion, the geometry that frames a pair of glasses — the face shape analyzer is the more stable read, and the glasses guide is where that stability actually pays off: frames sit on bone, and bone is the part that stays. If you want to know which measurements the pipeline is even taking, the explainer on landmarks and scores walks through it.

Frequently asked questions

At what age does a face scan score start to drop?

There is no fixed age, and no published threshold. What matters is which points a given tool uses. Tools that lean on eyelid and mouth contours tend to drift earlier, often from the late thirties. Tools that lean on bone landmarks can stay flat for decades. The change is gradual, not a step.

Can a face scan estimate my age accurately?

No. Age estimation from a single photo is a guess based on skin texture, folds and proportions, and it fails in both directions: it overestimates young faces in harsh light and underestimates older faces with smooth skin or a filter. Treat any age number as entertainment, not as a measurement.

Does losing weight change my score more than aging?

Usually yes, over a short window. A few kilograms change cheek fullness and jaw definition faster than a year of aging does. That is why a comparison without a weight log is close to meaningless.

Should I use the same photo for years to track change?

No. Use the same protocol, not the same file. Reusing an old photo freezes the camera, the light and the compression of that moment, and hides the very changes you are trying to see.