Is this image AI-generated? Four ways to check

1. Ask Google if its own watermark is inside

If the picture came from a Google model, it almost certainly has SynthID in it. You cannot see that signal yourself. Finding it needs the matching model. So this check goes through Google’s own tools.

Upload the image in the Gemini app and ask if it was made with Google AI. Google also runs a SynthID detector portal for journalists and researchers.

2. Look for content credentials

C2PA credentials are a signed note that travels with the file. They say what made it, when, and what was done to it after. When they survive, they are the most useful thing you will find.

Drop the file into a Content Credentials inspector. Or look for the small cr icon that some apps show next to an image.

Two big warnings. Most social apps strip metadata when you upload. And anyone can remove it in seconds. So missing credentials mean nothing at all.

3. Read the normal metadata

Before the clever checks, do the boring one. Any EXIF viewer will tell you if the file claims a camera, a lens, a shutter speed and a GPS location. Or if it only claims a software name.

This is weak evidence both ways. A real photo loses its EXIF after passing through most apps. And anyone can write fake EXIF into an AI image. But it is free and takes ten seconds.

4. Look at the picture properly

The old signs are mostly gone. Six fingers and melted ears are rare now. What is left is smaller, and still worth a look.

• Text. Shop boards, labels, book covers and number plates still break, especially away from the middle of the frame.
• Repeats. Crowds, leaves, railings and printed cloth often repeat too neatly, or turn to mush near the edges.
• Physics. Shadows pointing different ways. Reflections that do not match. A strap or a chain that passes through an arm.
• Everything equally sharp. A real lens focuses at one distance. If the foreground and the far hills are both crisp, be suspicious.

None of this is proof. All of it is a reason to do checks 1 and 2.

What about the “AI detector” websites?

Those are classifiers. They were trained on examples of AI images and real images, and they give you a percentage. They can help. They are also wrong very often, in both directions. They flag real photos and they clear AI images. They get much worse once a picture has been compressed, cropped or filtered.

So treat a percentage as one weak hint. A watermark check is evidence. A classifier is an opinion.

Putting it together

For a picture that really matters, do this in order.

1. Check for SynthID through Google’s tools. A match is close to settled.
2. Check for content credentials. Present and valid is strong. Absent tells you nothing.
3. Check the EXIF for a camera that makes sense.
4. Look at the text, the hands, the reflections and the focus.
5. Do a reverse image search. Finding the same picture on a stock site from 2019 beats every forensic check on this list.

And keep two questions apart. “Was this AI-made?” and “is this true?” are not the same question. Usually the second one is what actually matters.

Questions people also ask

How do I check if an image has SynthID?

Through Google’s own tools. Ask about the image in the Gemini app, or use the SynthID detector portal. There is nothing in the file you can read yourself, because finding it needs the matching model.

Does removing the visible watermark hide that an image is AI-made?

No. The sparkle, SynthID and content credentials are three separate layers. Taking the logo off changes how the corner looks, and nothing else.

Are AI image detectors accurate?

Not reliably. They are wrong in both directions often enough that a percentage should be treated as a hint, not a result, especially after the image has been compressed or edited.