What the reading actually measures

A smile score is not an opinion about your face. It is a measurement, and it is worth knowing exactly what goes into it.

Two muscles make a smile. The zygomaticus major runs from the cheekbone to the corner of the mouth and pulls the mouth upward — this is the part everybody can move on command. The orbicularis oculi is the ring of muscle around the eye socket; when it contracts it narrows the eye opening, lifts the cheek, and gathers the skin at the outer corner. Most people cannot contract it deliberately without also squinting, which is why it became the standard marker for separating a felt smile from a posed one. The distinction is named after Guillaume Duchenne, the 19th-century neurologist who mapped it.

How the number is built

Worth being straight about this, because the Duchenne framing invites an assumption the maths does not support. The score is a blend of two things:

  • A geometry reading (40%) — four landmark measurements, weighted mouth curvature 35%, eye constriction 25%, cheek raise 25%, symmetry 15%.
  • An expression classifier (60%) — a general happy-face confidence, not a Duchenne-specific one.

Multiply that through and eye constriction — the actual Duchenne tell — is about a tenth of the headline number, while mouth curvature carries more geometric weight than the eyes do. So the score answers “does this read as a happy face?” more than it answers “is this a Duchenne smile?” The Duchenne question is answered separately, by the four criteria below, and that is the part to read if authenticity is what you came for.

What each score band means

ScoreVerdictWhat it means
75–100Genuine JoyStrong on both halves — the expression reads unambiguously happy and the geometry agrees.
55–74Real SmileReads as genuine. The usual band for a normal, unposed photo.
35–54Polite SmileMouth is doing the work and little else — the social smile you produce for a camera.
0–34Fake SmileLittle happy-expression signal and weak supporting geometry.

55 is the genuine/not-genuine line. A photo can also clear it on expression confidence alone, which is why a 54 and a 56 are not meaningfully different — treat the band, not the digit.

The four Duchenne criteria

Separate from the headline score, four geometric checks run against your photo. Meeting three of four marks it as a Duchenne smile:

  • Eye engagementeye constriction at 60 or above. The orbicularis oculi tell — narrowed opening, gathered outer corner.
  • Cheek activationcheek raise at 60 or above. The lift that comes with a real eye contraction rather than a mouth-only smile.
  • Mouth expressionmouth curvature at 60 or above. The part everyone can do deliberately — necessary, nowhere near sufficient.
  • Facial symmetrysymmetry at 70 or above. Posed smiles skew more often than felt ones; the threshold is set higher for that reason.

Reading your own result

Treat the number as a description of one photograph, not of you. A smile is an event with a rise, a peak and a decay lasting well under a second, and a camera samples one instant of it. Photographed a third of a second early or late, the same genuine smile measures differently. This is the single most common reason a result feels wrong, and it is also why running two or three photos is more informative than running one — agreement across frames is the signal, and a lone outlier usually says more about shutter timing than about the smile.

What moves the number without your smile changing

Every landmark measurement inherits the limits of the image it was taken from, and a handful of ordinary photo conditions move a smile reading on their own:

  • Head angleEye constriction is measured across the eye opening. Turn far enough off-axis and one eye foreshortens, so the two eyes stop agreeing and confidence drops.
  • Flat, frontal lightingThe cheek lift and the fold at the outer eye corner are read partly through shadow. A direct on-camera flash fills exactly those shadows and can flatten a genuine reading.
  • GlassesFrames crossing the outer corner of the eye occlude the region the measurement depends on most. Heavy or thick frames cost the most accuracy.
  • Resolution and distanceEye-corner geometry is a small-scale feature. A face occupying a small share of a low-resolution frame gives the landmark detector less to work with than the mouth, so the eye half degrades first.
  • Recent expressionA held smile decays. Several seconds into holding one, the eye component relaxes before the mouth does — which is precisely the shape of a posed smile, even when it started out genuine.

What it cannot tell you

It cannot tell you whether you were happy. The Duchenne marker is a reliable correlate of felt emotion across populations, not a verdict on an individual — people do produce genuine smiles with little eye involvement, trained performers can produce the eye component deliberately, and some faces simply carry more resting tension at the eye than others. A low score means this photograph does not carry the geometric signature, which is a useful thing to know when you are choosing between shots for a profile or a headshot. It is not a statement about your sincerity, and no measurement taken from a still image could be.

It also cannot tell you whether people find you attractive. That is a separate question measured a separate way — by asking enough real raters and reporting where you land among them, which is what the 17-metric face score and the community Human Score do. Smile authenticity is one input to how a photo reads, not a ranking of the face in it.