Is this photo of a person real?

Fake profile pictures are the most common AI images you'll meet: dating profiles, marketplace sellers, sudden new followers, accounts that message you out of nowhere. Most can be caught in under a minute with the right checks. This guide gives you a repeatable process, from the fastest visual tells to the verification steps that settle it.

The stakes here are worth naming. A fake portrait is usually attached to a fake identity, a romance scammer, a bot account, a fraudulent seller, so "is this person real?" is often really "is this account safe to trust?" That's a reason to be thorough, and also a reason to be careful before accusing anyone: real people take unusual photos too. Aim for evidence, not a snap judgment.

Start with the parts AI still gets wrong

Look past the face itself, generators are excellent at faces now, and check the edges of the person instead. Ears and earrings often mismatch between sides. Teeth can blur together into a single ridge or repeat unnaturally. Glasses frames bend, or the two lenses have different shapes, or an arm of the frame vanishes into the hair. Hair strands dissolve into the background instead of crossing cleanly in front of it. Necklaces, collars, and shirt seams frequently break, change pattern, or fail to line up across the body. These peripheral details get less of the model's attention than the face, so they're where it slips.

Check the eyes closely

Eyes reward a zoom. In a real photo, the catchlights, the little bright reflections, match in both eyes because both catch the same light source. AI portraits often get two mismatched catchlights, or reflections that don't correspond to any light in the scene. Pupils can be subtly different sizes or shapes, and the fine detail of the iris can look smeared. The gaze itself sometimes has an unfocused, "looking through you" quality that's hard to place but easy to feel.

Check the background and the crop

AI portraits tend to have dreamy, out-of-focus backgrounds with melted details: text that isn't quite letters, windows that don't line up, foliage that has no real structure, other "people" who are shapes rather than persons. A suspiciously tight head-and-shoulders crop is itself a tell, it conveniently hides the hands and surroundings where mistakes live. Real profile photos usually include some honest, slightly messy context; a perfect floating headshot with a vague blur behind it deserves a second look.

One photo is a red flag by itself

Real people leave trails: multiple photos across time, different lighting, different angles, tagged appearances in other people's pictures, older posts that match. An account with exactly one flawless, studio-quality portrait and nothing else behaves like a generated identity, because producing a second consistent photo of the same invented person is still genuinely hard. Look for a photo history, not just a photo. Its absence is one of the strongest signals available, and it requires no image analysis at all.

Verify where the photo came from

Before you even reach for a detector, run the image through a reverse image search with Google Lens or TinEye. Two useful outcomes: either the photo turns out to belong to a real, unrelated person whose pictures were stolen, which exposes a catfish instantly, or it appears nowhere at all despite looking like a professional portrait, which is exactly what you'd expect from a freshly generated face. A stock-style photo that traces back to a real named individual is a different problem from a fake, but either way the account using it is not who they claim to be.

Then let a detector look

Eyes catch the obvious fakes; a detector catches the statistical ones. Run the photo through the detector, and crop to the face region for the cleanest read, a full screenshot with profile interface around it dilutes the signal and wastes resolution. A high AI score plus any of the visual tells above is about as close to certainty as this gets. If the score is ambiguous, lean on the non-visual evidence, the single-photo history, the reverse-search result, since those don't depend on the image fingerprint surviving.

Putting it together

No single check is decisive, but they stack well. The most convincing case combines a visual tell you can point to, a thin or nonexistent photo history, a reverse search that comes back empty or stolen, and a high detector score on a tight face crop. When those agree, you can act on it. When they disagree, treat the person as unverified rather than definitively fake, and let the safe course, not sharing personal information, not sending money, follow from the uncertainty itself.