Why Your Face Scan Score Changes Every Re-Take

One face, three different scores

Scan the same face twice and the score rarely matches. It is tempting to read the gap as meaning — a better day, a worse angle, a sign that the first number was right and the second was wrong. In practice the difference almost always comes from the input, not from you. A face scan is a ruler held up to a single photograph, and a photograph is the result of a dozen choices made in the second before the shutter.

The four variables that move the number

Most of the drift between scans traces back to a short list of physical variables:

• Distance and zoom. Move the camera closer and the face fills more of the frame; perspective widens the nose and shortens the jaw. Move it back and the opposite happens. Landmark detection reads those as real shape differences.
• Camera height. A lens at chin level and a lens at forehead level produce different jawlines, different forehead-to-eye spacing, and different cheek proportions from the same face.
• Light direction. Front light flattens features and hides contours. Side light carves a shadow the detector reads as a contour that is not there.
• Head pose and expression. A few degrees of tilt, or a smile on one side only, shifts every landmark at once.

Change any one of those and the score moves. Change two at the same time and you cannot tell which one did it.

Why the tool cannot separate you from your photo

A face scan has no way to know whether a number changed because your face changed or because the photo did. It measures the image it receives. That is not a flaw in the software — it is what landmark detection is. The honest reading of any face scan is "this is the geometry of this picture", never "this is the geometry of your face".

What the score is actually measuring

Under the surface, the tool locates a set of key points — eye corners, brow, nose tip, mouth corners, jaw outline — and compares the distances and angles between them. Every one of those points is estimated from pixels, and pixels depend on framing and light. That is why a scan is best understood as a camera check with a number attached, rather than a verdict about your bone structure.

How to get a repeatable reading

If you want a number you can compare over time, hold the camera variables still:

1. Same distance — about arm's length, not zoomed in.
2. Same lens height — at eye level, not above or below.
3. Same light — front-facing and even, with no strong shadow on one side.
4. Same expression — relaxed, mouth closed, eyes open normally.

Scan twice under those conditions before you believe any difference. If both scans land close together, the reading is stable. If they do not, something in the setup is still moving.

When a change is real

Repeatable scans taken weeks or months apart can reflect genuine change — weight, grooming, a haircut, a different pair of glasses. They can also reflect a new phone camera, a new room, or a new time of day. Before crediting the scan with anything, ask which variable you actually changed. Most of the time the honest answer is the lens, not the face.

A score is a photo, not a person

The most useful habit is to stop treating the number as a fixed property of your face. It is a description of one image, produced by one camera, in one patch of light. Once you accept that, the score becomes a tool for improving your photos — and it stops being a source of daily anxiety about a measurement no one else can see.

Frequently asked questions

Why do I get a different score every time I scan?

Because the photo is different every time. Distance, lens height, light direction, and expression all move landmarks, and the scan measures the photo rather than the person. Stabilize the setup and the score stabilizes with it.

Is the highest score my "real" one?

No. The highest score is simply the one produced by the most flattering combination of angle and light. It is not more true than a lower one — it is just a different photograph.

Should I re-scan until I get a good number?

If the goal is entertainment, take the number you like and move on. If the goal is a stable reading, re-scan with the same setup twice and compare those two, not your best-ever result.