Skip to the content
detectaicheck

AI image detector

Drop a picture and get a verdict built from four separate checks. Signed evidence counts the most and a classifier estimate counts the least, and the result always lists which checks ran. Use it to decide which images deserve a closer look, not as the final word on where an image came from.

Up to 10.0 MB. Kept in memory for 30 minutes at most, or deleted at once with Delete now.

What we check

  • Content Credentials

    We read the signed provenance record that some cameras, editors and AI tools attach to a file, check its signature and look for statements that an AI generator created or edited the picture, or that a device captured it.

  • Metadata traces

    We look through the EXIF, XMP, IPTC and PNG text fields for the names and settings that AI tools leave behind, such as a digital source type that says the picture was generated, or a saved prompt.

  • Invisible watermark

    We look for the hidden watermark that the reference Stable Diffusion code adds to every picture it creates. It cannot be seen, but it can be read back from the pixels.

  • AI image classifier

    A model trained on real photos and AI pictures looks at the pixels themselves and gives an estimate. It is the only check that still works when the file carries no other evidence.

How to check a picture

  1. Drop an image on the box above or pick one from your device. We accept JPEG, PNG, WebP, GIF, TIFF, HEIC and AVIF files. The watermark and the classifier read the pixels of JPEG, PNG, WebP and GIF files, the other types get the Content Credentials and metadata checks.
  2. Wait a few seconds while the checks run on our server.
  3. Read the verdict, then open the details to see each finding and which checks ran.

What a result means, and its limits

  • AI generated and AI edited come from evidence inside the file: Content Credentials, a watermark or a metadata trace. Likely AI is an estimate, mostly from the classifier.
  • No AI evidence found means that the checks which ran found nothing. It does not prove that a person made the picture, because metadata is easy to remove and many tools add no watermark.
  • The classifier misses many images from the newest generators and can sometimes flag a real photo.

Frequently asked questions

Can any tool tell for sure whether an image is AI?

No. Signed Content Credentials come closest, because the signature ties the claim to the file. Everything else is a clue. We show each clue with its strength so you can weigh it yourself.

Which generators can it spot?

Any generator that signs its pictures with Content Credentials or writes its name into the metadata, pictures from the reference Stable Diffusion code through its watermark, and the rest only through the classifier, which is the weakest check.

Why did a real photo get Likely AI?

Heavy filters, upscaling, strong compression or unusual lighting can confuse the classifier. Likely AI from the classifier alone is a weak signal, so look for a camera signature or other evidence before you draw a conclusion.

Does it work on screenshots?

A screenshot keeps the pixels but drops the Content Credentials and the metadata of the original, so only the watermark and the classifier still have something to work with.

Do you keep my picture?

No. The file is analyzed in memory and deleted after the result. The result page expires after 30 minutes.