Inspection report—
Waiting for an imageChoose one, or try an example.
AI labels (Content Credentials) stay with the image. Pixel watermarks like SynthID need Google's SynthID Detector.
Content Credentials checker
Drop in an image to see whether it carries C2PA Content Credentials, which company issued them, and whether they say AI was involved.
Checked in your browser. Nothing is uploaded. Turn off your Wi-Fi and it still works.Check an image for Content Credentials
JPEG · PNG · WebP · up to 25 MBor drop it here, or paste it anywhere on the page
How it works
Short answer
Content Credentials are a signed record attached to an image under the C2PA standard. They can say which app or AI model made or edited it. Drop an image above to see whether it has them, which issuer is named, and whether it's marked as AI-generated.
Images: three layers, three answers
Metadata is where your location and camera details hide, so that's what the cleaner removes. Content Credentials stay, because removing an AI label would mislead whoever sees the image next.
What the checker looks for
Content Credentials live in different places depending on the format: a JUMBF box in JPEG files, a caBX chunk in PNGs, and a C2PA chunk in WebP. The checker reads each of them and reports:
- Whether credentials are present. A signed manifest is attached to the file.
- Who issued them. OpenAI, Anthropic, Google, Adobe, Microsoft, Samsung, Leica and others are recognised by name.
- Whether AI is declared. The IPTC term trainedAlgorithmicMedia means fully AI-generated, and compositeWithTrainedAlgorithmicMedia means partly AI-edited.
To verify the signature itself, drop the same file into the Content Authenticity Initiative's verify tool.
It also checks the image's ordinary metadata for AI labels and embedded prompts, since many tools write those even when they don't sign anything.
Why credentials go missing
Content Credentials are metadata, so they only survive if every app the image passes through keeps them. Taking a screenshot, re-saving in a basic editor, or uploading to a platform that strips metadata removes them. That's why Google and OpenAI also add SynthID, a watermark hidden in the pixels themselves, which survives where metadata doesn't. Read SynthID explained for how the two fit together.
Cleaning without losing the label
If you use the clean button here, location, camera and date details are removed but Content Credentials stay in place. We think an AI label should travel with the image it describes.
Questions
What are Content Credentials?
A tamper-evident record, built on the open C2PA standard, that travels inside a media file. It can list the tool that created the image, edits made afterwards, and whether generative AI was used. The record is cryptographically signed by the company that issued it.
Which tools add Content Credentials?
Images from ChatGPT and the OpenAI API, Google's Gemini app, Adobe Firefly, Microsoft's Image Creator, and a growing number of cameras and phones such as the Pixel 10, which signs every photo. Coverage keeps widening as more companies join the C2PA.
Does this check verify the signature?
Not yet. We find the credentials, read the issuer and the AI label, and show them. To verify the signature itself, use the Content Authenticity Initiative's free tool at verify.contentauthenticity.org.
No credentials were found. Is the image real?
Not necessarily. Screenshots, many social networks and most editing apps strip credentials. A missing record only means there is no record.