What this page does, and what it does not
It checks the picture: whether the file says a generator made it, whether it carries the records a camera writes, and how much it looks like the AI pictures a detector model learned from. It cannot tell you whether the person exists or is who they say they are, and no picture check can. A real person's stolen photo passes every test here, because it is a genuine photo.
How to use it
- Add the photo. Use the original file if you can: a screenshot or a copy from a messenger has lost its metadata.
- Read the headline and the reasons. "AI-made: the file says so" is solid for that file. "Looks like a real camera photo" means camera records are present. A score is only an estimate.
- Then do the rest of the routine in our guide to checking a profile photo: search for the same photo elsewhere, read the account around it, and ask for a live video call.
Why faces need extra care
Faces get the best-trained generators, and the old giveaways (mismatched earrings, melted glasses, garbled text) have been fading. Our own test had only five real photos of people and five faces from an early face generator: too few for a rate, enough to say that a score on a face is less certain than a score on a landscape. The detector as a whole flagged 85% of AI pictures and 25% of real ones, and it was wrong most often on drawings and paintings.
Before you use a result
- Do not accuse anyone. A result describes a picture file. A real person can have an AI-looking or AI-edited photo, and an AI face can pass every check.
- Do not post a score as proof. If you report a profile, report what you saw: "the file carries a prompt from a generator", or "the photo appears on another site under a different name".
- Only add photos you may check. The check runs in your browser, but like every tool on this site the page also sends a copy of each picture and its result to our server (see the privacy policy). Do not add someone's photos if that would not be right.
What the file can show
| Finding | What it points to |
|---|---|
| A prompt, workflow, Content Credentials or an AI label naming a generator | The file was made by an AI generator. Certain for that file. |
| The invisible Stable Diffusion watermark | The same, found in the pixels. Rare, because many front ends never add it and a JPEG save removes most of it. |
| Camera records (make, lens, settings, exact time) | A photograph, probably. Can be copied; a photo can be edited afterward. |
| A score from the detector | How AI-like the picture looks. Right about 85% of the time on AI pictures and 75% on real ones in our test, and less sure on faces. |
| Nothing in the file | Nothing. Messengers and social sites strip the data from real photos too. |
Frequently asked questions
Does this recognize the person in the photo?
No. It does not identify or match faces. It looks at the picture file and its pixels, and a result is about the file only.
The photo passed. Is the profile real?
Not necessarily. A photo of a real person can be used by someone else, and it passes every picture check. Search for the photo elsewhere and look at the account's history and what it asks of you.
It flagged my own photo as AI. Why?
The detector flags about a quarter of real pictures, most often retouched, smooth or stylized ones. Open the reasons: if the file carries camera records, that points the other way. A score is not a verdict.
Can I use a result to report a fake profile?
Report what you observed, such as a prompt found in the file or the same photo under another name. A detector score alone is not something a platform can check, and it should not be used to accuse a person.