AI Image Detector
Is this picture AI generated? A verdict, a score and the evidence.
The file's own metadata is read on your device. The image is then sent to a server to be examined, held in memory for a second and never kept.
Drop an image here
Find out whether it was AI generated. Paste works too.
Reading the file…
That didn't work.
This tool reads the file on your device, then sends the image to a server to be examined. It is held in memory for the second the request takes, never written to disk, and never kept afterwards.
Keyboard shortcuts
- / or ⌘K
- Search every tool
- ⌘↵
- Run this tool, from anywhere on the page
- ⌘S
- Download the result
- ?
- Open and close this list
About the Pathine AI Image Detector
Drop in a picture and get a verdict on whether it was AI generated, with a score out of 100 and the evidence it rests on. It reads the file first and looks at the pixels second, and it tells you which of the two decided the answer.
The file is the better witness. DALL·E, Firefly, Imagen and Midjourney embed Content Credentials that name the generator; Stable Diffusion front ends leave the whole prompt in the PNG; Google and Adobe stamp an IPTC flag that says trained algorithmic media. A camera signs the other way, with a make, a model and an exposure. When any of that is present the verdict is close to certain, and it is found on your device before anything is sent.
Then the image goes to a model that examines it the way a forensic analyst would: hands, teeth and lettering, lighting that disagrees with itself, textures that repeat, the waxy over-clean look, and the sensor noise and clutter that real photographs have and generators do not. It lists what it saw and gives a probability. That half is an opinion, and the page shows it as one.
When you'd reach for it
A photo in a listing or a profile that looks too good
Rental listings, marketplace sellers and dating profiles are where generated images do the most quiet damage. Check the picture before you trust the person.
An image going into a story, a report or a slide
If it turns out to be generated after publication, the correction costs more than the check. Two seconds here is cheaper than a retraction.
Something a student or a contributor handed in
A submitted illustration or photograph that should have been made by hand. The file's own metadata is often the decisive part, and it is shown in full.
A viral picture you are about to share
The image that is too perfect a fit for the argument usually is. Check it, then share it or do not.
How to check whether an image is AI generated
-
Drop in the image
Or paste it. JPG, PNG, WebP, HEIC and AVIF, up to 8MB.
-
Read what the file says first
That appears immediately. Content Credentials, generator settings, a camera record, or none of those — most images from social platforms arrive stripped bare, and the page says so rather than treating an empty file as innocent.
-
Then the model's read
A second later: what kind of image it is, up to six things it noticed and which way each points, and a probability with a confidence attached.
-
Take the verdict for what it is
When the file declares itself, trust it. When only the pixels are speaking, treat the score as a strong opinion — enough to ask a question, not enough to accuse anyone.
Common questions
How accurate is it?
Two different answers. When the file carries a declaration — Content Credentials from a generator, a Stable Diffusion prompt, an IPTC flag — the verdict is as good as the file, which is very good. When it is judging from pixels alone, it is a well-briefed guess: right most of the time on the generator look, and least reliable on polished professional photography and on images that have been resized, screenshotted or compressed several times. A photoreal scene from the newest generators with no people, hands or lettering in it can pass as a photograph. The score tells you which situation you are in.
Why does it say there is no metadata?
Because nearly every platform removes it. Instagram, X, WhatsApp, Facebook and most messaging apps strip Content Credentials and EXIF on upload. A file with nothing in it is the normal case, not a suspicious one, and the page never counts an empty file as evidence either way.
Can it be fooled?
Yes. Metadata can be removed or faked, though nobody forges a manifest that says their photograph was made by DALL·E. The pixel read can be beaten by a good generator, a careful edit, or heavy compression. A low score is not a certificate of authenticity, and a high one is a reason to look closer, not a conviction.
What are Content Credentials?
A signed record inside the file, in a format called C2PA, saying who made it and what was done to it. OpenAI, Adobe, Google, Microsoft, Midjourney and several camera makers write them. This tool reads what the record says; it does not verify the cryptographic signature, so it reports the claim as a claim.
Does it detect Photoshop's generative fill?
When Adobe's own metadata survives, yes: it is recorded as a composite with trained algorithmic media, and the page reports AI edits rather than an AI image. The underlying photograph's camera record usually survives too, which is why a camera name on its own is never treated as proof.
Is my image uploaded?
The file is read on your device first, and that part works with no connection at all. The pixel read needs a model too large to run in a browser tab, so the image is sent to a server, held in memory for the second the request takes, and never written to disk or kept.