How to Tell if a Photo Is AI-Generated: Signs You Can Spot by Eye

People ask how to tell if a photo is AI-generated for two different reasons. Some want to avoid getting fooled by a fake, while others want their own generated image to slip past unnoticed. Either way, the tells are the same.

Sign one: hands and fine details

Fingers, knuckles, the way a hand actually grips an object. Models have gotten noticeably better at this over the past year, but zoom in and it still gives itself away.

Same story with zippers, buckles, straps, and glasses temples. Anything thin that has to connect to something else.

Sign two: text

A label looks fine from a distance. Zoom in and the letters turn into shapes that only resemble letters.

Short words sometimes land. A full sentence almost never does. Cyrillic text breaks down even faster than Latin.

Sign three: reflections and light

A glossy surface reflects something that isn't actually in the scene.

A shadow falls the wrong way relative to the light source, or there's no shadow at all.

The lighting is almost too even: no awkward shadow, no stray highlight.

How to tell an AI photo by texture

Skin looks smooth to the point of resembling a mannequin, with no pores and no texture.

Fabric reads as painted rather than woven. The pattern is too clean, with no natural skew.

A repeating print breaks down on curves. Plaid and stripes drift instead of staying aligned.

Sign five: too clean

There's no clutter in the frame. No crumb, no speck of dust, no fingerprint smudge.

A real photo always carries something extra and unplanned, and that's exactly what separates it from a render.

What about metadata

Some tools embed a tag in the file marking it as AI-generated. That tag usually survives a crop, but re-saving the file through an editor often strips it out.

Metadata alone isn't something to rely on. Its absence proves nothing either way. We go deeper on this in a separate piece: how platforms detect AI content through metadata.

If you need a generated image that actually works

Every sign above gets fixed the same way: ask for imperfection on purpose. Uneven light, a bit of grain, everyday clutter, visible skin texture.

And check the image at full zoom, not in a thumbnail preview. That's exactly where half these problems show up.

On metadata, honestly, I'm not sure how closely platforms actually check it. I've heard conflicting claims and haven't seen a confirmed case either way.

FAQ

What are the clearest signs a photo is AI-generated? Hands and small mechanical details, text that falls apart up close, reflections that don't match the scene, lighting that's too even, and skin or fabric texture that looks too smooth.

Can you tell from the file's metadata? Sometimes. Some tools embed a tag marking the file as generated. But its absence proves nothing, and re-saving through an editor usually removes it anyway.

Why does Cyrillic text break down more than Latin text? Cyrillic shows up far less often in training data than Latin text, so models end up reproducing letter shapes instead of actual letters. That pattern shows up mainly with Cyrillic, not with Latin text.

How do you keep a generated image from getting called out? Ask for imperfection: uneven lighting, a little grain, random objects in frame, real skin texture. Then check the result at full zoom before you use it.

Sources

1. Flami: how platforms detect AI content through metadata
2. Flami: why AI models mangle hands and text
3. Flami: why Cyrillic trips up image generators