Image provenance checker
Is this image AI?
Drop an image and we read the provenance signals it carries: signed Content Credentials, labels in its metadata and known invisible watermarks. We name each signal at its true strength and say what we could not check. This is not a pixel-based AI image detector; those guess.
Checked in your browser. Your image is never uploaded.
Check an image
Drop an image hereChoose an image to check
Or paste one with Ctrl + V (Cmd + V on a Mac).
JPEG, PNG, WebP, AVIF, HEIC or GIF, up to 50 MB, one image at a time.
Checked in your browser. Your image is never uploaded. How it works
Checking
What we found
What we checked
The Content Credentials check did not finish, so no signature or trust result is shown. Your file was still read for metadata.
Where to check further
Some signals can only be checked by the companies that made them. These links open their own sites in a new tab. We never send them your image or our results; if you use one, you upload the image to them yourself.
Technical details
How to read the result
We report what a file carries and how far each signal can be relied on. We never say that an image is or is not AI-generated, because these signals cannot show that.
-
Verified
A signed Content Credentials record that checks out. Its signature is valid, the image has not changed since it was signed, and the signer is on the C2PA Trust List. It tells you who signed which claims. It does not prove the claims are true.
-
Verified on the Interim Trust List
The same checks, but the signer appears only on the older Interim Trust List, which was frozen on January 1, 2026. In our tests, files from Adobe Firefly and Microsoft Image Creator land here.
-
Signed, signer not trusted
The signature is valid and the image matches the record, but no trust list vouches for the certificate that signed it (usually because it is on neither list; sometimes because it has expired), so the name it carries is not verified. Anyone can make a certificate with any name. Genuine OpenAI files land here too, so this tier cannot tell a genuine record from one made with a self-made certificate. It does not mean the claims are true.
-
Declared
Labels in unsigned metadata, such as IPTC or XMP fields, and records that failed validation. Anyone can add, change or remove them. A label, not proof.
-
Decoded
A known invisible watermark found in the pixels. We look for the pattern that Stable Diffusion's reference software adds. Anyone running that software adds it, and it is not a signed record. Ordinary JPEG compression, resizing and cropping remove it, so it mostly survives only in original, unshared files.
-
Hint
A tool name appears somewhere in the file's metadata. It is a hint, not a declaration of how the image was made.
-
Nothing found
Most images lose these signals when they are shared, screenshotted or re-saved. Finding nothing does not mean the image is real or human-made.
Visible marks, and what each one tells you
Some AI tools put a visible mark on their images. Here is what the common ones look like and what each can and cannot tell you. We never look for these marks ourselves: you look at the image, and a mark is a lead rather than proof, because anyone can add one or remove it.
-
The four-pointed star
- Looks like
- A small four-pointed star in a corner of the image.
- What is reported
- It is the visible mark of Google's Gemini app. Reported: since August 14, 2026 people can switch it off (Settings, then Media Watermark), and the invisible SynthID watermark and Content Credentials stay on.
- What it tells you
- Where present, a label from the Gemini app, not a signed record. Without it, nothing can be said.
-
"Imagined with AI"
- Looks like
- A short text label in a corner of the image.
- What is reported
- Meta says it adds this label to images made with its own AI tools.
- What it tells you
- A label from Meta's side. Text can be added to any picture or cropped away. If you can see it, tick the box below to add Meta's checker to your links.
-
"AI-generated content"
- Looks like
- The words "AI-generated content" at the bottom left of the image.
- What is reported
- Reported on images edited with Samsung's Galaxy AI features. Samsung is also reported to add Content Credentials to AI-edited images, which this checker reads.
- What it tells you
- A label, not a signed record. Check the file's Content Credentials as well.
-
The row of coloured blocks
- Looks like
- A short row of small coloured blocks in a lower corner.
- What is reported
- Reported on images from DALL-E 2. We found no documented visible mark for OpenAI's newer image tools.
- What it tells you
- At most a hint about DALL-E 2 images. It is easy to crop off, and its absence says nothing.
-
The small "b"
- Looks like
- A small letter b in a corner of the image.
- What is reported
- Reported in 2023 on images from Microsoft's Bing image tools.
- What it tells you
- A small mark that is easy to miss, crop or imitate. Treat it as a lead.
The drawings are ours, made to show the idea. Position, size and look vary by product and version. More detail on visible AI watermarks.
Learn more
- How to tell if an image is AIWhat can be checked, what cannot, and where.
- What is SynthID?Google's invisible watermark, and who can check it.
- What are Content Credentials?Signed records, trust lists and their limits.
- Visible AI watermarksThe marks you can see, and what they prove.
- Privacy and methodWhat runs in your browser and what is never sent.