How to check an image for AI labels and Content Credentials
What is stored inside an image file, how to read it, and how far you can trust it.
What is it?
A reader for the information hidden inside an image file
AI Image Credential Checker opens an image file and reads the provenance information stored in it: the record of how the image was made that many cameras, editors and AI image tools now write into their files.
You normally cannot see this information. It sits beside the picture data, and most apps do not show it. The checker finds it, translates it into plain words, and says whether the file claims to be AI-generated.
The file is read in your browser. It is not uploaded, and there is nothing to sign up for.
What it reads
Four kinds of information a file can carry
The checker looks for each of these and shows whichever it finds.
Content Credentials
A signed record defined by the C2PA standard. It names the tool that made the file, the actions taken, and a source type such as generative AI. Adobe, Google, OpenAI and several camera makers write it.
IPTC source type
A standard metadata field that says how the image was produced. The value trainedAlgorithmicMedia means it was created by a generative AI model.
Tool and camera details
The software or camera named in the file, the date, the author and copyright fields, and whether a location is stored.
Generation settings
Stable Diffusion and ComfyUI save the prompt and settings as text inside PNG files. If they are there, the image came from one of those tools.
Background: the C2PA specification, the IPTC Digital Source Type vocabulary and IPTC's guidance for AI-generated media. If you sell on Google Shopping, note that Google Merchant Center requires AI-generated product images to keep the IPTC trainedAlgorithmicMedia tag.
Reading the result
What each result means, and what it does not
The checker gives one of five headline results. Each is a statement about the file, not a judgment of the picture.
| Result | What was found | What it does not prove |
|---|---|---|
| This file says it was made with AI | An AI source type in its Content Credentials or IPTC metadata, or saved generation settings. | That the label is genuine. This page does not verify signatures. |
| Its metadata names an AI image tool | The software or credit field names a known AI image tool, with no formal AI label. | How much of the image the tool made. Some tools also edit real photos. |
| It carries Content Credentials, with no AI label | A C2PA record that does not mark the image as AI-generated. | That no AI was used. Check the record with the Verify tool. |
| Its metadata describes a camera or editing software | Ordinary camera or software details and no AI markers. | That it is a real photo. These fields are easy to edit or copy. |
| No credentials or provenance information found | Nothing about how the image was made. | Anything at all. Most images online have had this information removed. |
How to use it
Four steps to read a file
The page opens with a small sample that carries an AI label, so you can see a full result first.
- Get the original file. Download the image itself. A screenshot, or a copy saved from a chat app, has usually lost its information.
- Open it here. Choose the file, or drop it on the page. The result appears at once.
- Read the headline and the list. The headline is the summary. The list shows each thing found, and the details box shows every field that was read.
- Verify if it matters. If the file has Content Credentials, open the official Verify tool and check the same file there to confirm the signature.
Limitations
What it cannot do
- It does not verify signatures. It reads Content Credentials but cannot confirm who signed them or whether the image was changed afterwards. Use the Verify tool for that.
- It cannot detect AI from the picture. It never analyses the pixels, so an AI image with no metadata looks the same to it as a photo with no metadata.
- It cannot read invisible watermarks, such as Google's SynthID. Those need the maker's own detector.
- Metadata is easy to remove. Uploading to most social platforms, sending through chat apps, taking a screenshot or re-saving in an editor can strip everything.
- Metadata is easy to fake. Anyone can write a camera name or an AI label into a file. Only a valid signature is hard to forge.
- Some records cannot be read. Compressed Content Credentials, and records stored online instead of in the file, are detected but not opened.
- AVIF and HEIC support is basic. Those files are searched for known markers, not fully parsed.
Using it fairly
A label is evidence, not a verdict
Do not accuse someone of faking an image because a file has no credentials, and do not treat a camera name as proof that a photo is real. Both mistakes are common.
For anything that matters, such as news, a dispute or a purchase, combine what the file says with other checks: where the image first appeared, who published it, and whether the official Verify tool confirms its record.
If you publish AI images yourself, keep their labels. Stripping an AI label to pass an image off as a photo is deceptive, and some platforms require the label to stay.
Check your first image
The checker at the top of this page starts with a labelled sample. Open your own file to see what it carries. Preparing product photos? Run them through the Product Photo Checker too.