What gives away your AI photo? Look past the face

Your AI photo rarely gives itself away in the face itself, but in the details around it: earrings that don't match, teeth that read as one white block, glasses that don't rest on the ears, or background text that means nothing. Modern models now render faces so convincingly that people can barely tell them from real ones. The weak spot has moved to the edges of your image, and that's exactly where viewers are getting sharper.

Why the face is rarely the weak spot

The face is usually not where an AI photo breaks, because that's what the models are trained on most. In research from UNSW Sydney (published in the British Journal of Psychology, 2026), regular participants spotted an AI face only 50.7% of the time, no better than a coin flip; even people with an exceptional memory for faces got no further than 57.3%.

AI faces are, if anything, too average: strikingly symmetrical and "typical", without the small irregularities of a real face. Someone judging your photo rarely consciously notices that the face is right, but they do trip over a detail that isn't. So aim your check at everything around it.

Jewelry: earrings, necklaces and rings that don't add up

Jewelry is the most common giveaway, because the model keeps paired and fine objects poorly consistent. What to watch for:

• Earrings that aren't the same shape or size on the left and the right.
• A necklace that disappears into the skin or clothing halfway instead of lying over it.
• Rings that don't actually wrap around the finger, or a watch whose strap doesn't continue.

Half or floating jewelry stands out faster than you'd think. Keep accessories deliberately simple in your prompt, or leave them out if they add nothing to the image.

Teeth, ears and glasses: the small anatomy that shows

Alongside hands, teeth, ears and glasses are where AI most often slips up. Teeth sometimes become one even white block with no separation, or there seem to be a few too many. Ears differ in shape between left and right, or an earlobe is missing.

Glasses don't sit neatly on the ears, or the arms don't continue behind the head. A wide smile and a pair of glasses always reward a second look, because those are exactly the spots viewers subconsciously double-check.

Background, text and reflections: where the model stops paying attention

The subject gets the model's attention, the background gets its mistakes. Text on signs, packaging or name badges is a classic: scrambled letters, invented characters or just slightly misspelled words. Reflections in windows, mirrors or wet streets often don't match the subject in front of them.

Worth knowing: the familiar tips "count the fingers" and "look for garbled text" are getting less reliable, because newer models increasingly get hands and short text right (per Malwarebytes). The giveaways have shifted toward jewelry, reflections and paired details instead.

How to check and fix the details before you publish

Before you publish, run a short checklist and repair targeted slips instead of regenerating the whole photo.

1. Zoom in on ears, teeth, jewelry and hands, and on any text in the frame.
2. Compare paired items left and right: earrings, eyes with their catchlights, glasses arms.
3. Fix a small slip with inpainting in the AI Photo Editor: select just the earring or the tooth and regenerate that piece.
4. If the error runs deeper, re-render in the photo generator with a simpler description of the accessories, or crop out the problem area.

In short: one wrong detail can give away an otherwise convincing photo, while a minute of checking prevents it.

Frequently asked questions

Can you still spot an AI photo by the face?

Usually not. In the UNSW study, regular viewers sat at 50.7%, coin-flip level. The face is typically the strongest part of an AI photo; the details around it give the image away sooner.

What is the most common giveaway?

Paired and fine objects: earrings that don't match, a necklace that vanishes into the skin, and text in the background. Those are the fastest to check.

How do I fix such a detail without remaking everything?

With inpainting. You select only the problem area, for example one earring or the teeth, and regenerate that part. The rest of the photo stays untouched.

Want your AI photos to hold up down to the details? Create an account and pay only per render you use.