How to spot AI-generated images

AI image generators have improved fast, but they still leave fingerprints. If you know where to look, many fakes fall apart under thirty seconds of attention. Here are the tells that catch most of them.

1. Hands and fingers

Hands remain the classic giveaway. Count the fingers, then look at how they bend: AI images often show six fingers, joints that fold the wrong way, or two hands merging where they touch. Rings, watches, and held objects frequently melt into the skin.

2. Text and logos

Look at any writing in the image: signs, labels, book covers, shirts. Generators tend to produce letter-like shapes rather than real words, and brand logos come out slightly wrong. If the text looks like it is written in a dream, that is a strong signal.

3. Backgrounds and edges

The subject usually gets the most model attention, so the background is where quality slips: railings that do not line up, windows with impossible reflections, crowds of half-formed faces, and objects that blend into each other. Follow straight lines across the image and see if they stay straight.

4. Lighting and shadows

Check where the light comes from. AI images often mix light sources that do not agree: a face lit from the left with shadows falling the wrong way, or reflections in glasses and eyes that do not match the scene in front of the person.

5. Skin, hair, and texture

AI skin tends toward an airbrushed, waxy smoothness, and individual hair strands often merge into painterly clumps. Fabric patterns are another good check: stripes and checks that warp or change scale mid-garment are hard for generators to keep consistent.

6. Symmetry that is slightly off

Earrings that do not match, glasses with two different frames, eyes with different catchlights, collars that sit unevenly. Humans are roughly symmetric, and generators often lose track of the pair.

7. The gut check, then a detector

If something feels off but you cannot name it, that instinct is data. Run the image through an AI detector for a second opinion, and crop to the suspicious region, a hand, a face, a patch of background, so the detector scores exactly the part that worried you.