How to read the report
The tool above is the job: one clear face, then a report in a fixed order. Read the range before the integer.
The number and the likely range
You look sits over one integer: the best apparent age for this photo, not a birth date. Likely range runs from a low year to a high year. A span wider than 20 years is reset to 10 years either side of the integer. A span under 2 years is opened to 1 year either side. The ends then stay inside 1 to 100, so a result near those edges can finish slightly tighter. An integer outside 1 to 100 is dropped, not shown.
Confidence, cues, and a hard stop
Confidence is low, medium, or high. Low still shows the integer and adds a note that photo quality limited the estimate. High means the face was readable, not that the year is a chronological age. Why this number lists 3 to 5 cues: a feature name, one observation, and Reads older, Reads younger, or Neutral. Tips to look younger lists 2 notes on light, grooming, and the next photo, not a treatment. The closing line says this is apparent age in this photo, not for age verification or medical advice.
No face, more than one face, or features that cannot be made out at all each get their own message. A soft but readable photo still returns a report, with confidence low and the limit written into a cue.
Why the range is the result
Mean absolute error averages absolute gaps, so a three-year average can hide a ten-year miss. NIST IR 8525 says that figure does not count large overestimates, and is the wrong metric for age verification. Keep the years where two ranges overlap.
- Read Likely range before the integer under You look.
- If confidence is low, change the photo before you change your mind.
- Reads older describes this picture. It is not a diagnosis.
What published tests actually measured
On 30 May 2024 NIST published Face Analysis Technology Evaluation results for age estimation and verification (NIST news release): six algorithms, about 11.5 million photos (visa files, FBI mug shots, border webcams, immigration photos from more than 100 countries of birth). The 2014 visa test, about 6 million photos, asked only for an age. This run also asked whether someone was over 21. NIST recommends no particular use. On the shared visa photos, mean absolute error fell from 4.3 years to 3.1 years, and five of six algorithms beat the best 2014 result. Error still moves with image quality, gender, region of birth, and age, and was almost always higher for female faces. NIST says the reason is unknown.
On Mexican visa photos, NIST IR 8525 finds ages 0 to 17 more accurate than adults. The leading algorithm misses by 0.34 years under age 1, about four months, and by 3.5 years at age 30. That is not this page's score. A phone selfie adds blur, a close lens, filters, cosmetics, and retakes.
Where the miss gets larger
Error also depends on the age in the photo. Ofcom's 29 October 2026 assessment, for a possible UK rule at 16, cites NIST: for most algorithms mean absolute error rises with age. The lowest figures Ofcom quotes are 1.77 years at ages 13 to 17 (19 of 50 algorithms under 3 years) and 2.39 years at ages 18 to 25 (9 of 50 under 3). That set is far larger than May 2024's six (Ofcom's rapid assessment at 16). A border webcam differs from a mug shot under two lamps, and a beauty filter strips texture. Low confidence flags that photo.
Curiosity is not an age check
Age assurance is how a service enforces an age rule. Ofcom treats facial age estimation as one method that can be highly effective, beside photo-ID matching and banking checks, because it returns only an approximate age. Anyone under a challenge age, as shops use Challenge 25, must pass a second method. Ofcom's October 2026 assessment names four methods it judged capable at 16: photo-ID matching, digital identity, banking checks, and facial age estimation. A low average still hides how often a system overestimates age badly.
How Old Do I Look AI checks no ID, returns no pass or fail, and sets no challenge age. The report is for curiosity, not age verification or medical advice. Do not use it to buy a restricted product, open an account, or judge someone else. Upload your own photo, or a photo of someone who asked.
Phone, laptop, and a second photo
On a phone, Take a selfie opens the camera. On a laptop, drop, paste, or choose a JPG, PNG, or WebP up to 25 MB. The browser shrinks the long side first. You get 10 estimates a day, no account. The photo is resized on your device, then analyzed in memory and never stored (privacy policy).
Next time change only the light, the distance, or the expression, and keep the years the ranges share. Close-selfie distortion: the face age calculator. An old print: the age detector. Friends: guess my age. Someone else: someone else's age. How the cues are produced: the AI age guesser. A chatbot on the same file: how old do I look ChatGPT. The main test: How Old Do I Look AI.
Age estimator questions
Is the number my real age?
No. It is how old the face appears in that photo. A gap of several years either way is ordinary.
What does mean absolute error mean?
It is the average size of the miss, high or low. A three-year average is not a promise about your photo. Read the range on the report.
Is an age calculator by photo the same as counting from a birthday?
No. A birthday calculator subtracts a date you type. It never sees a face. An age calculator by photo only estimates appearance, and a picture cannot recover a birthday.
Can this prove I am over 18?
No. Where facial age estimation is used for a legal rule, Ofcom also expects a challenge age and a second method. This page returns neither a pass nor a fail.
Why do two photos disagree?
Light, distance, expression, and filters change the file. Keep the overlap of the ranges. The face age calculator covers the close-selfie effect.
Do you store the photo?
No. The photo is analyzed in memory and never stored. It is not written to disk, a database, or logs. See the privacy policy.