AI image generators have gotten startlingly good. The fakes that once announced themselves with melted faces now pass a casual glance — and they are being used for scams, fake product listings, and viral misinformation. You do not need special software to protect yourself. You need to know where generated images still slip up, and build two or three verification habits. Here is how.

Start with the hands — then check teeth and text

Hands remain the classic tell. Count the fingers: AI models still produce six fingers, fused digits, or hands that melt into sleeves and coffee cups with disturbing regularity. It is the first thing to check because it fails so often.

Next, look at teeth and mouths — too many teeth, teeth fused into a single white strip, or smiles that do not quite align with the face. Then look for any text in the image: signs, labels, book spines, t-shirt slogans. AI gibberish text — letters that look right at a glance but spell nothing — is one of the most reliable giveaways still standing.

Study the light and shadows

Real photographs obey physics; generated ones approximate it. Check whether shadows fall in consistent directions and match their light sources. Look at reflections in mirrors, windows, and sunglasses — AI often renders reflections that show something entirely different from the scene, or faces reflected at impossible angles. Earrings, glasses frames, and jewelry are worth a close look too: one earring different from the other, or glasses arms that do not connect, are common glitches.

Backgrounds betray fakes

Image generators lavish attention on the main subject and get sloppy with everything else. Zoom into the background: faces in crowds that smear into each other, architectural details that warp and repeat, staircases that lead nowhere, fences with inconsistent slats. If the background looks like a dream you half-remember, that is a strong signal.

Skin and texture: too perfect is suspicious

AI portraits often have a distinctive over-smoothness — skin like polished plastic, with no pores, blemishes, or stray hairs. Real cameras capture noise, slight motion blur, and imperfect focus. An image that is razor-sharp everywhere, with cinematic lighting and flawless skin on every face, deserves skepticism rather than admiration.

For video: watch how things move

Deepfake videos have their own tells. Watch for unnatural blinking — too rare, too regular, or absent. Look at the edges of the face and hair for flickering or shimmer as the head turns. Lip-sync is another weak point: in a fake, the mouth movements often lag the audio slightly or look oddly smooth during fast speech. And be wary of videos where the person never turns their head far to the side — many face-swap techniques break down at sharp angles, so scammers keep the subject facing forward.

Verify, don’t just eyeball

Visual inspection catches a lot, but the strongest defense is verification habit:

  • Reverse image search. Drag the image into Google Images or TinEye. If the “photo” of a breaking news event actually appeared months ago in a different context — or appears on stock and AI-gallery sites — you have your answer.
  • Check the source, not just the image. Who posted it? A brand-new account with no history pushing an outrageous claim is a red flag regardless of how real the picture looks. Established outlets with editorial standards are still the best filter.
  • Look for provenance info. An increasing number of cameras and AI tools attach Content Credentials (the C2PA standard) — tamper-evident metadata saying how an image was created. Some platforms now display a small badge for it. Its absence proves nothing (screenshots strip it), but its presence is meaningful.
  • Be extra skeptical of emotional content. Fakes spread because they provoke — outrage, pity, awe. If an image makes you feel something strong and demands you share it right now, that urgency is the actual payload. Pause.

What to do when you’re not sure

Do not share it. That is the whole rule. Sharing an uncertain image to ask “is this real?” still spreads it to everyone who will not read your caption. If it matters — a scam targeting your community, a claim about a real person — report it on the platform and, for potential fraud, tell the person being impersonated through a channel you trust.

The bottom line

No single trick catches every fake, and the tells will keep shifting as the technology improves. But the combination works: check hands, teeth, and text; interrogate light and backgrounds; distrust perfection; and verify the source before the image. The goal was never to become a forensic analyst — it is to become the person in the group chat who pauses for ten seconds before hitting share.

Curious how the technology behind these images works? Our generative AI explainer breaks it down without the jargon.