Free metadata checker · C2PA and IPTC

Check AI metadata: what is inside your image file?

I read the provenance data that AI tools write into image files: the C2PA manifest and the IPTC value DigitalSourceType. You see where the evidence sits, what the file says about tool and date, and which EU label would fit.

Your images stay on your device.
Drag images here
JPEG, PNG, WebP, AVIF · 1 to 20 images
What I read in the file

How I read the metadata.

  1. 01The C2PA manifestA signed record inside the file, known as Content Credentials: which tool, which action, when. I show who issued it according to the certificate. Whether the signature is valid, I do not check.
  2. 02IPTC DigitalSourceTypeA plain-text value in the XMP block. trainedAlgorithmicMedia stands for an image created with generative AI, compositeWithTrainedAlgorithmicMedia for an existing image edited with generative AI.
  3. 03Where it sits and the labelFor each image you see whether the evidence sits in the manifest, in the IPTC field or in both, plus tool, issuer, date, action and the raw value. From this I suggest AI GENERATED or AI MODIFIED.

What evidence is and what it is not

EvidenceMarker in the original fileThe tool that created or edited an image writes it in, with tool, action and date. That is all I look for.
HintA detector's estimateA detector calculates a probability from the pixels. The image check here uses no hints.
No AI evidenceNo marker, no statementI found no marker. That says nothing about how the image was made.

Unreadable means: the file cannot be read, for example because it is damaged or the extension does not match the content.

Frequently asked questions about the metadata checker

More background in the Guides, for example C2PA and Content Credentials: how to check what is in an AI image

General information, not legal advice.

Which metadata does the checker read?

The C2PA manifest, known as Content Credentials, and the IPTC value DigitalSourceType in the XMP block, in JPEG, PNG, WebP and AVIF. I show whether the evidence sits in the manifest, in the IPTC field or in both. I do not display camera or location data from EXIF.

What do trainedAlgorithmicMedia and compositeWithTrainedAlgorithmicMedia mean?

They are IPTC values for the origin of an image. In IPTC terms, trainedAlgorithmicMedia means “Created using Generative AI”; for that I suggest AI GENERATED. compositeWithTrainedAlgorithmicMedia means “Edited using Generative AI”: an existing image that a generative AI model has augmented, corrected or enhanced; for that I suggest AI MODIFIED. In the measurement of 1 September 2026, Canva wrote this value for an image made from one’s own photo with an AI background.

What does the checker show for a piece of evidence?

Where it sits, the tool according to the file, the issuer according to the certificate, the date, the action (created or edited) and the raw value from the file. As the tool I show what the file states: for images from Google, that is the signing library, not a model name.

Do you verify the signature?

No. I read the C2PA manifest and show who issued it according to the certificate. Whether the signature is valid, I do not check.

Why do I find no evidence in my file?

Not every tool writes it: on 1 September 2026, 34 of 47 measured models in a generator catalogue wrote a marker. And the export path can delete it: the same generation carried a C2PA manifest as a native PNG, none as a JPEG from the web interface, and a screenshot kept no provenance data. No AI evidence only means: I found no marker.

Does the checker read Content Credentials from the cloud?

No. In the measurement of 1 September 2026, Adobe Firefly only wrote a reference to an online manifest into the file. I display the reference, but I do not resolve it. The image then shows as no AI evidence.

Does the checker detect invisible watermarks?

No. I only read the metadata of the file; I do not assess the pixels. So I do not see a watermark hidden in the pixels, and no AI assesses anything here.

Does the metadata survive in a Shopify store?

On the stored original, yes: after upload, all 34 markers were kept on 1 September 2026. On the copies Shopify delivers in the store, only 6 of 38 marked images still carried the marker on 30 September 2026. So check the original file as it comes out of the tool.

Are my images uploaded?

No. I read the files here in your browser; they do not leave your device. The report is created on your device too.

How do I label an image with evidence?

In your store with the app: I check the originals in your Shopify admin and suggest the label, you accept it, and a theme block shows the EU label. Free plan always free with 25 labels, 7-day free trial of Unlimited. For ads and marketplaces, the free AI Label Generator puts the label straight into the image; the copy keeps the IPTC data of the original.

Label the whole store?

In the app I find the AI evidence in your product images, you accept with one click, and the official EU label appears in the store.

View on the Shopify App Store