UtilityNest

AI Image Checker

Find out whether a picture was made or edited with AI by reading the tags hidden inside the file. The checker reads C2PA Content Credentials, the signed record that ChatGPT, Adobe Firefly, Google's Gemini image models and Microsoft's image tools attach, and it looks for unsigned AI metadata such as the IPTC digital source type and the generation settings Stable Diffusion and ComfyUI save in PNG files. Everything is read in your browser; the image is never uploaded.

This checker reads tags stored in the file. It cannot see invisible pixel watermarks such as SynthID, which Google and OpenAI also add, and it does not guess from how a picture looks. For those, try Google's SynthID Detector, the Gemini app or openai.com/verify. A result of "No AI tags found" is not proof that a person made the image. See all of a photo's metadata

How to use AI Image Checker

  1. 1Drop one or more images on the page. JPG, PNG, WebP, AVIF and HEIC all work, up to 20 at a time.
  2. 2Wait a moment the first time while the Content Credentials reader loads. Later checks take well under a second.
  3. 3Read the verdict at the top of each card: Made with AI, Edited with AI, AI tags found in the metadata, Content Credentials with no AI step, or No AI tags found.
  4. 4Open the details to see which app signed the file, who issued the signature, when it was signed, every recorded step, and whether the signature still matches the pixels.

What the checker reads

There are two kinds of hidden tag in AI images, and the checker reads both. The stronger one is a C2PA manifest, also called Content Credentials. It is a small signed record stored inside the file (in an APP11 segment of a JPEG, or a caBX chunk of a PNG) that lists the app that made the image, the steps it went through and a digital source type such as trainedAlgorithmicMedia, which is the IPTC term for an image created by a generative AI model. The record is signed with a certificate, so the checker can tell you who signed it and whether the image has been changed since. We read it with the C2PA project's own open-source library, compiled to run inside your browser.

The weaker kind is ordinary metadata that anyone can write or edit: an IPTC or XMP digital source type field, a creator or software field naming an AI tool, or the generation settings that Stable Diffusion's AUTOMATIC1111 interface saves in a PNG text chunk called parameters and that ComfyUI saves as prompt and workflow. These are good hints but not proof, because they are not signed.

Which AI tools leave tags

According to their own documentation, OpenAI adds both C2PA Content Credentials and Google's SynthID watermark to images made with ChatGPT and its API; Adobe applies Content Credentials to images generated entirely in Firefly; Microsoft applies them to images from its Azure OpenAI image models and the Microsoft 365 Copilot app; and Google embeds C2PA metadata in images from its Nano Banana Pro model in the Gemini app, alongside SynthID. Meta says its AI images carry invisible watermarks and metadata. Stable Diffusion and ComfyUI write generation settings by default unless the user turns that off. Checked on October 30, 2026; vendors change these practices often, and our guide How AI image generators hide tags in pictures links to each source.

What each result means

Made with AI means a signed C2PA step says the picture was generated by a trained model, even if a later app (Photoshop, for example) edited it afterwards. Edited with AI means a signed step says part of the image was made with AI, such as generative fill on a real photo. AI tags found in the metadata means unsigned fields name an AI tool or contain its settings. Content Credentials, no AI step listed means the file is signed, for example by a camera or an editing app, and nothing in the record mentions AI. No AI tags found means the file carries none of these tags.

The signature line tells you whether the record still matches the picture. If it says the file was changed after signing, someone edited the pixels without updating the credentials, so the history can no longer be trusted for this copy. A separate certificate note (for example an expired or test certificate) is about the signer, not about changes to the image.

Why "no AI tags" doesn't prove a person made it

Metadata is fragile. OpenAI's own help page says Content Credentials can be lost through everyday actions such as taking a screenshot, and uploads, conversions and messaging apps often remove them too. That is why OpenAI and Google also hide SynthID, a watermark woven into the pixels themselves. This checker cannot read SynthID, because the detector is not public. To check for it, upload the image to Google's SynthID Detector or ask the Gemini app whether Google AI made it, or use OpenAI's checker at openai.com/verify for OpenAI images. The checker also does not guess from how a picture looks, so it will never call a real photo AI just because it looks too perfect.

What happened when we tested it

We checked eight files. An Adobe Firefly image from the C2PA project's open-source test files came back as Made with AI, naming Adobe Firefly as the generator, Adobe Photoshop 26.9 as the app that later cropped and edited it, Adobe Inc. as the signer, and a signature that still matched. A camera-style C2PA test image with only colour edits came back as Content Credentials with no AI step. PNG files we saved with the Stable Diffusion parameters chunk and with ComfyUI's prompt and workflow chunks, and a JPEG with the IPTC trainedAlgorithmicMedia tag, were all flagged as AI tags in the metadata, with the prompt, sampler and model shown. A plain photo returned No AI tags found. All eight were read in about 1.2 seconds, including loading the reader, and the page made no network requests with the images.

Frequently asked questions

How can I tell if an image was made with AI?

Drop it on this page. If the file still has its Content Credentials or AI metadata, the checker shows which tool made it and when. If it finds nothing, the tags may have been removed, so also try Google's SynthID Detector or the Gemini app, which look for an invisible watermark instead.

Can it detect ChatGPT images?

Yes, when the file still has the Content Credentials OpenAI adds. Images downloaded straight from ChatGPT carry them; a screenshot or a re-saved copy usually doesn't. OpenAI also adds a SynthID watermark, which this page can't read, but openai.com/verify can.

What are Content Credentials (C2PA)?

An open standard for a signed record inside a file that says which app made it and what was done to it. It's backed by Adobe, Microsoft, Google, OpenAI and others, and the record shows the cr icon in apps that support it.

Does the checker upload my image?

No. The file is read in your browser with the open-source C2PA library compiled to WebAssembly. The page also turns off the library's optional online lookups, so nothing about the image is sent anywhere.

Why does it say No AI tags found for an image I know is AI?

Most likely the tags were removed by a screenshot, an upload to a social site or messenger, or a format conversion. Some generators also add no tags at all. The result only describes what is in this copy of the file.

Can it read Google SynthID?

No. SynthID is hidden in the pixels and Google's detector is not public. Use the SynthID Detector portal or ask the Gemini app about the image.

What does "the file was changed after signing" mean?

The pixels no longer match the hash stored in the Content Credentials, so someone edited the image without updating its record. The listed history describes an earlier version, not necessarily this one.

Can it tell a real photo is real?

Only when the photo carries Content Credentials from a camera or phone that signs its pictures, such as some Leica, Nikon and Sony models or Google's Pixel 10 camera. Most photos have no such record, so the checker shows No AI tags found for them.

Related guides