AI Image Detector — Was This Picture Made With AI?
Upload one image. DiffALL reads what the file declares about where it came from, asks a detector model what it sees, and gives you a number from 3 to 97 with every reason behind it.
Free, no sign-up — your file is deleted as soon as the check finishes.
Drop an image here or click to browse
JPG · PNG · WebP · BMP — up to 20 MB
Free, no sign-up — your file is deleted as soon as the check finishes. The answer is an estimate between 3 and 97, never 0 or 100.
Reading the file…
What DiffALL measures
Content Credentials, verified
ChatGPT, Gemini, Adobe Firefly, Bing and Microsoft Designer sign what they make (C2PA). DiffALL reads the signature and checks it against the Content Credentials trust list, so you see who signed it and whether the file has changed since.
AI metadata and generator settings
The IPTC “trained algorithmic media” tag Google and Meta write, and the prompt-and-sampler settings Stable Diffusion, ComfyUI, InvokeAI and NovelAI leave in the file.
A detector model for the rest
Most files lose their metadata on the way to you. Then a vision model trained on thousands of generators estimates how likely the picture is to be AI-made.
Never 0, never 100
Even a signed record only proves what the file says, and no model proves anything. The number stops at 3 and at 97, and the middle of the scale is a real answer: not conclusive.
Every reason shown
What the file records and what the model saw are shown separately, so you can see which one the number came from.
What it cannot see, said up front
A small AI edit inside a real photo is almost invisible to any whole-image detector. The page says so rather than calling such a photo clean.
Supported formats
Use cases
Before you share it
A dramatic news photo in a group chat: check whether it carries a generator's signature, and what a model makes of it.
Moderating uploads
Get a first signal on submitted images, with the reason, before a person looks.
Check a stock or client image
See whether a file declares AI generation before you license it or put your name on it.
Know what your own export says
Find out whether the AI tool you used labelled the file, and whether that label survived your edits.
Frequently asked questions
How does DiffALL tell whether an image is AI-generated?
In two stages. First it reads what the file itself declares: signed Content Credentials (C2PA), the IPTC “trained algorithmic media” tag, generator settings left by Stable Diffusion-style tools, or a Software field naming an AI generator. If the file says it was made with AI, that is the answer. If it says nothing, which is most of the time, a detector model estimates the likelihood from the pixels.
Why is the number never 0 or 100?
Because neither would be true. A signed record proves what the file says, not what happened, and a model's estimate is an estimate. So the scale runs from 3 to 97. Below 25 we found no signs of AI, from 85 the signs are strong, and in between we cannot tell either way. In our tests, among images scored 85 or more, about 9 in 10 really were AI-made.
Can it detect an image that was only partly edited with AI?
Only when the file records it, through Content Credentials or AI metadata. A small region replaced by AI inside a real photograph is hard for any whole-image detector to see: in our tests, when AI replaced a tenth of a real photo, the result reached the “strong signs” range fewer than one time in eight. So a low number never means “not edited”.
My photo is real but scored high. Why?
Detectors make mistakes, most often on digital art, heavy filters, very small pictures and screenshots of screenshots. That is why the result shows its reasons and why the top of the scale stops at 97. If you know the truth, tell us under the result — that is how we find where it goes wrong.
Does it work on WhatsApp or social-media copies?
Those copies lose their metadata, so only the model can answer, and compression removes some of what it reads. It still works on most of them, but less reliably than on the original file.
Is my image stored?
No. It is deleted as soon as the check finishes, and checking the Content Credentials never sends anything to another service.