AI Image Checker
Drop in an image to see what it says about how it was made: signed Content Credentials, AI labels, saved prompts and the camera or app named in it.
How the AI Image Checker works
It reads four things from the file, all in your browser:
- Content Credentials. A signed record defined by the C2PA standard, read with the official open-source c2pa-web library. It checks the signature, follows the record back through earlier steps, and looks the signer up in the public C2PA trust list.
- The IPTC source type. A standard label for how an image was made. “Trained algorithmic media” means generative AI.
- Generator settings. Some image generators save the prompt and settings in the PNG file. The checker recognises the formats of Stable Diffusion web UIs, ComfyUI, InvokeAI and NovelAI, and shows the prompt.
- Camera and software fields. EXIF and XMP name the camera, the app that saved the file and, sometimes, the GPS location. An AI tool’s name in these fields is flagged.
It doesn’t look at the pixels or guess. Each finding is shown with where it came from, and whether a signature covers it.
What the results mean
- Made with AI: a valid signed record says the image was made by generative AI. This is the strongest finding.
- Edited with AI: a signed record says part of the image was made by generative AI, for example a generative fill.
- Labelled as AI: unsigned metadata says AI, such as an IPTC label, a saved prompt or an AI tool named as the software. Usually right, but anyone can add or remove it.
- Credentials don’t match this image: the file has a record, but the image or the record changed after it was signed. Don’t rely on what the record says.
- Signed as a camera capture or Has Content Credentials, no AI label: a valid record lists how the image was made, and none of the steps is labelled as AI.
- No AI labels found: the file says nothing either way. That is what you’ll see for most images from social networks, messaging apps and screenshots, AI or not.
A GPS location finding is a privacy note: anyone you send the original file to can see where it was taken.
Frequently asked questions
Can this tell for sure whether an image was made by AI?
No, and no tool can from the file alone. The checker reads what the file says about itself: its Content Credentials, its IPTC source-type label, generator settings and the software named in its metadata. A signed Content Credentials record is strong evidence of how the file was made. Unsigned labels anyone can add or remove. And an image with no labels at all may still be AI: most sites and apps strip metadata when an image is uploaded or shared, and many generators never add any.
What are Content Credentials (C2PA)?
A standard from the Coalition for Content Provenance and Authenticity for a signed record inside an image file. The record lists the app that made the image, the steps taken, such as created, edited or cropped, and the images it was built from, and it is signed with a certificate. If the image or the record is changed afterwards, the signature no longer matches, and the checker says so.
Which AI tools add Content Credentials?
Some do. OpenAI, for example, says images made with ChatGPT and its API carry a C2PA record. Others add none, or only an unsigned label. Even when a generator adds them, saving a screenshot, re-encoding the image or uploading it to most social networks removes them, so a missing record tells you little.
What does “not signed” mean?
The label sits in ordinary metadata (EXIF, XMP or PNG text) that any photo editor can write. It’s a useful hint, and the checker shows it word for word, but it isn’t proof: it can be added to a real photo or removed from an AI image.
What is the trust list?
Anyone can make a certificate and sign a record with it. The C2PA keeps a list of certificate authorities that check who they issue certificates to, and the Content Credentials project keeps an earlier list used by cameras and apps. When an image’s signer comes from one of them, the checker shows “Signer on trust list”. A signer not on the lists can still be genuine, but the name in the certificate hasn’t been vouched for.
Does it detect invisible watermarks such as SynthID?
No. Invisible watermarks are hidden in the pixels and can only be read with the maker’s own detector. This checker reads the file’s metadata and signed records, not the pixels.
Is my image uploaded?
No. The image is read in your browser and never sent to Samprix or anyone else. The page downloads the C2PA reading library and the public trust lists, which contain nothing about your image.