Is This Photo AI-Generated? A Free Step-by-Step Check

You saw a photo and something felt off — a face that’s a little too smooth, a background that doesn’t quite hold together, hands that won’t survive a second look. Or maybe it looks completely normal and that’s exactly what worries you. “Is this photo AI-generated?” is now one of the most common questions on the internet, and the honest answer is: you can narrow it down quickly, but your eyes alone won’t settle it.

This guide walks through the checks you can do yourself in under a minute, explains the trap that fools most people, and shows how to confirm a suspicion for free.

1. Start with the eyes, hands and teeth

Generators have improved, but fine, high-frequency detail is still where they slip. Look closely at:

Eyes — pupils that aren’t quite round, two eyes with mismatched reflections, or a glassy, lifeless stare.
Hands — extra or merged fingers, knuckles that bend wrong, jewellery that fuses into skin.
Teeth — rendered as one smooth white block instead of individual teeth.
Ears and accessories — earrings, glasses frames, and hairlines that warp or dissolve.

These are the same weak points that reveal a face swap, and they’re a good first pass.

2. Check the background and edges

The subject often gets the model’s attention while the background falls apart. Scan for text that turns into nonsense glyphs, repeated or smeared patterns, windows and tiles that don’t line up, and objects that melt into each other. Then check the boundary between the subject and the scene — AI images frequently have a faint halo or an unnaturally clean cut where a person meets the background.

3. Inspect lighting and reflections

Real scenes obey one set of light sources. Generated ones often don’t. Check that shadows fall in a consistent direction, that reflections in glasses, mirrors and water match what should be reflected, and that the catchlights in both eyes agree. A subject that looks “pasted into” the lighting is a strong tell.

Five high-yield zones to inspect first: edges, eyes, hands, fine text, and lighting consistency.

The reverse-image-search trap

Here’s the mistake that catches almost everyone. You run the photo through Google Lens or a reverse-image search, it finds nothing, and you conclude the photo must be genuine.

It’s the opposite. Reverse search is good at one thing: finding where an existing photo has appeared before. A freshly generated AI image has never existed anywhere — so reverse search returns nothing, and “no results” gets misread as “it’s real.”

This is why the two approaches are complementary, not competing: reverse search for stolen real images, forensic analysis for synthetic ones.

Why the eye test isn’t enough

Every visual check above is real — but each one is beatable. Models are trained specifically to pass them at a glance, and once a photo has been through social-media compression, the subtle texture and colour cues your eye relies on are flattened away. A still that looks completely clean to you can still carry forensic traces in data you can’t see.

That’s the gap automated analysis fills. Instead of judging what the image looks like, it measures what the image is made of:

Layer	What it looks at	What an AI image tends to show
AI model score	Learned generation patterns	High generation probability
Frequency (FFT)	Spectrum of the pixels	Periodic grid fingerprint
Error Level Analysis	Compression consistency	Edited regions light up
Metadata / C2PA	Origin data	Missing camera EXIF, AI provenance tags

How to check a photo for free

You don’t need to read spectrums yourself. To verify a suspicion:

Save the highest-quality copy of the image you can (screenshots lose detail — get the original where possible).
Upload it to FakeRadar’s free analyzer.
Read the result as signals, not a verdict. FakeRadar reports whether AI-generation signals were detected and how strongly — across an AI model score, frequency and error-level analysis, a dedicated face-swap check, and metadata — rather than a black-box “fake / real” stamp.

Reading signals instead of a single label is the whole point. It’s also why detection is signal-based: the value is in how independent layers agree, not in any one number.

FAQ

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

No. Modern generators are built to beat the obvious tells, and re-compression erases fine texture. The eye test catches clumsy fakes; the reliable signals live at the pixel level.

Does a reverse image search prove a photo is real?

No. A freshly generated image has never existed online, so reverse search returns nothing — which people misread as proof. Reverse search finds stolen real photos; it’s blind to synthetic ones.

Is checking a photo for AI free?

Yes. FakeRadar analyzes an image for AI-generation signals for free, no account needed for a first check, and shows you the evidence rather than a yes/no.

What’s the single most reliable signal?

There isn’t one. Reliable detection stacks independent signals so a weak result on one layer is backed by others.

Summary

Start with eyes, hands, teeth, backgrounds, edges and lighting — the cheap, fast checks.
“Reverse search found nothing” is not evidence a photo is real — for AI images it’s expected.
Your eyes can’t see compression and frequency traces; forensic analysis can.
Stack independent signals and read the result as signals, not a verdict.
Check any photo for free when a manual pass leaves you unsure.