ImageGuesser

Real or AI face? How to tell if a portrait is AI-generated

Faces are what image models are best at, because they have seen more faces than anything else. That makes a fake portrait the hardest kind of image to call, and the most common one in scams. The giveaways are no longer in the face itself but around it.

Why it is hard to tell

Modern image models have been trained on enormous numbers of portraits, and the obvious errors of a few years ago, like melted features or mismatched eyes, are now rare. A generated face is usually symmetrical, well lit and plausible. Looking at the center of the face and asking "does this look like a person?" will almost always give you the wrong answer.

What the models still get wrong is consistency. A face is surrounded by things that have to agree with each other: two ears, two earrings, two eyes reflecting the same light, a hairline that has to meet a background. Each of those is predicted separately, and that is where the seams show.

What gives it away

Catchlights that disagree

Zoom into both eyes. In a real photo, the small reflections of the light source have the same shape and position in each eye. Generated eyes often reflect different windows or no light at all.

Teeth as a single strip

Count the teeth. AI teeth often blur into a uniform band, have too many front teeth, or lose the gaps and slight irregularities real teeth have.

Ears and earrings

Compare the left and right ear. Mismatched earrings, earlobes that merge into the jaw, and ear folds that make no anatomical sense are still common.

Hair that dissolves

Look where hair meets the background or shoulders. Strands that fade into the backdrop, or a hairline that turns into texture, point to generation.

Glasses and collars

Glasses frames that change thickness or bend into the skin, and shirt collars that are different on each side, are paired-object failures.

Skin with no history

Real skin has pores, small marks, stray hairs and uneven color. A face that looks retouched to perfection in a casual photo deserves suspicion.

Quick checklist

Before you guess, run through these checks:

  • 1
    Compare the light reflections in both eyes.
  • 2
    Count the teeth and look for uniform blurring.
  • 3
    Check that both ears and both earrings match.
  • 4
    Follow the hairline into the background.
  • 5
    Check glasses, collars and other paired items for symmetry.
  • 6
    Look past the face: is the background readable and consistent?

Where you will run into them

Fake profiles
Generated faces are popular for fake social media, dating and professional-network accounts because, unlike stolen photos, they do not show up in a reverse image search.
Romance and investment scams
Scammers use consistent sets of generated portraits to build a believable persona over weeks of conversation.
Fake reviews and testimonials
A smiling face next to a glowing review is cheap to generate and adds false credibility.

How to verify it

Run a reverse image search with Google Lens or TinEye. A stolen photo usually turns up elsewhere under a different name. A generated face often returns no matches at all, which is suspicious in itself for an account that claims a long online history.

Look at the whole profile, not one photo. Real people have photos taken in different places, by different people, with different cameras and light. A set of images where the person always looks the same and always looks perfect is a warning sign. When it matters, ask for something a generator cannot easily fake on demand, like a live video call.

If you suspect Google's models were used, upload the image to the Gemini app and ask whether it was made with Google AI. It checks for Google's invisible SynthID watermark.

Frequently asked questions

Can you tell if a face is AI-generated just by looking?

Sometimes, but not reliably. Modern models produce convincing faces, so look for consistency failures around the face, such as mismatched catchlights, earrings, ears and hairlines, and combine that with a reverse image search and a check of the rest of the profile.

Why does a reverse image search find nothing for a fake profile picture?

Because a generated face is new. It has never been posted anywhere else, so there is nothing to match. For an account claiming years of history, no matches at all is a warning sign.

Do AI-generated faces still have wrong hands and eyes?

Much less often than before. Current models usually get the face itself right. The errors have moved to paired details like ears, earrings and eye reflections, and to the boundary between hair and background.

Test your eye on today's puzzle

Six images, one real photograph. Every fake is made with Google's latest image models, and a new puzzle goes live every day.

Play the daily challenge