How old do I look? We built the estimator and turned it off.

What our test found

The estimator was built to our strictest privacy standard: landmarks are computed in the browser, geometry only, no image ever uploaded. On a 30-photo validation set (ages 6–85, stratified) the best model had a median absolute error of 13 years: an 85-year-old could read as 22 or 63, a 6-year-old as 18. A hand-tuned heuristic did worse (median 27). The strongest single signal, brow width to face length, correlated well in fitting but collapsed on new faces. The full numbers are in our method notes: fitting-set median 7 years, validation median 13: classic overfitting, and the reason the tool is not on this page.

The conclusion generalises: the main carriers of age (wrinkles, skin laxity, pigmentation, texture) are texture signals, not geometry. Reading them means a different kind of model, a different privacy story, and a different honesty budget. We're not willing to fake it in the meantime.

What actually makes a face look younger or older: the changeable list

If you came here to look younger (or just to understand what people read when they guess your age), these are the factors that carry the signal, roughly in order of how much you control them:

• Sun exposure. The single largest controllable factor. Photoaging (spots, roughness, deep lines) is overwhelmingly UV damage; daily SPF is the best-evidence intervention on this list.
• Sleep. Chronically short sleep shows up as paler skin, darker under-eye circles and drooping eyelids within days, and reads as years in perceptual studies.
• Expression lines and resting face. Repeated folding plus collagen loss etches lines; how much you frown, squint and purse matters, as does whether you sleep face-down.
• Hydration and skin barrier. Dehydrated skin looks duller and more lined than it is; this one is visible within hours, in both directions.
• Smoking and alcohol. Both accelerate wrinkling and sallowness measurably; quitting reverses part of the visible damage within months.
• Teeth. Tooth shade and visible wear are strong, heavily weighted age cues, and one of the most reliably changeable.
• Brows and hair. Sparse or greying brows and thinning, faded hair add perceived years; a fuller brow and a deliberate hair color are fast, reversible levers.
• Weight and posture. Subtle facial fullness loss with age is a real cue, but posture and bearing change perceived age immediately.
• Expression and energy. A neutral or tired resting face reads older than the same face animated, which is why strangers' guesses swing with your mood.

The honest summary: most of what reads as "age" is skin, hair, teeth and expression: nearly all of it changeable, none of it measurable from facial geometry. That's why a geometric age number would have been theater, not information.

What about AI age estimators elsewhere?

Texture-based models can guess age better than our geometric experiment: that's exactly what they're for. But they need to see your skin up close, they age badly across skin tones, and any site that runs one on an uploaded photo now holds a biometric-adjacent image of you. If you try one, prefer tools that process the photo on your device and publish their error range. If a site shows a single number with no error bar and no method note, treat the number as entertainment.