How to Tell If an Image Is AI-Generated by Eye
How to tell if an image is AI-generated: a visual checklist for hands, text, reflections, light and backgrounds, and what newer models get right.
- Artifact categories in the Kamali et al. research guide
- anatomical, stylistic, functional, violations of physics, sociocultural
- Newer-generator images our main detection engine caught
- 41/42 (Sightengine, tested 2026-10-13)
| Where to look | What older generators often got wrong | Newer generators | How much weight to give it |
|---|---|---|---|
| Hands and fingers | Extra or fused fingers, joints bending the wrong way, hands melting into what they hold | Much less common; still worth checking in crowded scenes and small background figures | Low: a normal hand clears nothing |
| Text and lettering | Gibberish on signs, labels, book spines and T-shirts | Big headline text is often clean; small or secondary text can still drift into nonsense | Medium for small print, low for large clear text |
| Reflections and mirrors | Reflections that don't match the object; a mirror showing a different room | Improved, but reflections in glasses, windows and water can still disagree with the scene | Medium |
| Light and shadows | Shadows falling in different directions; a face lit from a side with no light | Mostly coherent; mismatches show up more in scenes with several light sources | Medium when the mismatch is clear |
| Backgrounds and crowds | Repeated faces, windows or tiles; people merging into each other | Less common; look at the far background and the edges of crowds | Medium |
| Perspective and geometry | Straight lines that bend, stairs that lead nowhere, railings that merge | Still worth checking in buildings and interiors | Medium |
| Everyday objects | Bikes without chains, chairs with odd legs, keyboards with random keys, earrings that differ | Improved, but small mechanical details are a common slip | Medium |
| Skin, hair and texture | Waxy, airbrushed skin; hair dissolving into the background | Often very realistic; phone beauty modes produce the same smooth look | Low: real photos look like this too |
| Context and plausibility | Wrong uniforms, impossible signs, events that never happened | A generator cannot know what really happened, however good the pixels are | High when you can check it against the real world |
Can you tell if a picture is AI just by looking?
Sometimes, but not reliably any more. You can learn to spot a lot of AI images by eye, yet newer generators have fixed many of the classic mistakes, and real photos have oddities of their own: motion blur, lens distortion, panorama stitching errors, portrait-mode blur cutting into hair.
So if you want to know how to tell if an image is AI-generated just by looking, treat what you see as clues, never as a verdict. A 2024 research guide by Kamali and colleagues sorts the tell-tale signs into five groups (anatomical, stylistic, functional, physics and sociocultural) and stresses that human-visible artifacts are not always present. Use the checklist above to collect clues: one odd detail is not proof of AI, and a flawless picture is not proof of a camera.
Faces, hands and bodies
Count fingers, then look at how the hands hold things: a cup handle passing through a finger, a phone with no grip. Check ears, teeth and the two sides of a pair of glasses or earrings. In group photos, look past the main subject to the people at the edges and in the back, where mistakes are more likely to survive.
Text, logos and small print
Read everything you can: street signs, shop fronts, labels, the newspaper in someone's hand. Large headline text is often clean now; the small print is where letters still blur into shapes that look like writing but aren't. Logos that are almost right, with a letter missing or a shape slightly off, are another sign worth noting.
Light, shadows and reflections
Find the main light source and check that shadows fall away from it consistently, including under people's feet. Then look at anything reflective: windows, mirrors, sunglasses, puddles, the eyes. Do the reflections show what should be there, from the right angle? These are physics problems, and generators still sometimes paint a plausible-looking reflection that doesn't match the room.
Backgrounds, patterns and geometry
Zoom into the background. Look for faces that repeat in a crowd, windows that change size along a building, tiles or fence posts that merge, stairs and railings that don't connect, and lines that should be straight but wobble. Ordinary objects are a good test too: count the spokes and the chain on a bicycle, check that chair legs reach the floor, and look closely at keyboards and jewellery.
Does the scene make sense?
This is the check that does not get weaker as generators improve. Ask whether the whole scene is plausible: the right uniform for that police force, the right plugs and road signs for that country, a public figure at an event they didn't attend, weather that doesn't match the date. A picture can be flawless pixel by pixel and still show something that never happened.
How to tell if an image is AI-generated when your eyes are not enough
If the clues are mixed, or you need more than a hunch, run the picture through our AI image checker and follow the four-step check. In our test, the detection model behind it caught 41 of 42 images from newer generators. It isn't perfect either, and it works best on the original file rather than a screenshot (why screenshots are harder).
Frequently asked questions
What's the quickest way to tell if an image is AI-generated by eye?
Check the hands, the small text and the reflections, then ask whether the scene makes sense. If nothing stands out, that doesn't clear the image; it only means the quick checks found nothing.
Do AI images still have six fingers?
Sometimes, but newer generators get hands right far more often than early ones did. Don't rely on it.
Are AI images always too smooth or too perfect?
Often, but phone beauty modes and editing apps produce the same look on real photos, so smooth skin is weak evidence either way.
Can I trust a picture that passes every check here?
No. Passing the visual checks only means you didn't find a mistake. Look for Content Credentials, run a detection model, and check the context against the real world.
Is there one trick for how to tell if an image is AI-generated?
No. Every visual tell has exceptions, and generators keep fixing the ones people talk about. Combine several clues, check the context, and use Content Credentials and a detection model for anything that matters.