AI Image Checker

Check any image for signs of AI generation and see why. This AI image checker reads the file's credentials and metadata and runs a detection model locally, so the picture never leaves your device.

Drop a image here, paste with Ctrl+V, or choose a fileTap to choose a image

JPG, PNG, WebP, AVIF, HEIC · runs in your browser, nothing is uploaded

An AI image checker that shows its work

Most tools return a single percentage and leave you to guess what it means. An "87% AI" label does not say whether the generator signed the file, whether a prompt was found in the metadata, or whether a model simply thinks the pixels look synthetic. Those are very different levels of certainty, so this AI image checker keeps them apart:

The verdict at the top follows the strongest evidence available. The list of reasons below it tells you exactly what was found.

How to use the AI image checker

  1. Drop an image onto the box, click it to choose a file, or press Ctrl+V (Cmd+V on a Mac) to paste a copied image.
  2. To check an image on a web page, copy the image address and paste it into the link field. Some sites block other pages from reading their images; in that case save the image and drop the file instead.
  3. Wait a moment. The first check downloads the detection model (about 22 MB) and stores it in your browser, so later checks are almost instant.
  4. Read the verdict, then open "Show details" to see the signer, software, camera data and exact model score.
  5. Press "Check another image" to clear everything and start over.

What each AI image check result means

Credentials found with an AI step

The file contains a C2PA manifest whose actions include a "trained algorithmic media" source, or a composite with AI-generated parts. ChatGPT, Gemini, Adobe Firefly and Microsoft Designer downloads usually look like this. You will see the name of the tool and the organisation that signed the claim.

Credentials found without an AI step

Some cameras and editors sign every image they save. A signed capture from a camera with no later AI step is good evidence for a real photo, as long as the signature validates.

Generator named in metadata

Fields such as Software, CreatorTool or the IPTC Digital Source Type mention a generator, or a PNG file carries a Stable Diffusion prompt or ComfyUI workflow. Metadata is not signed, so someone could fake it, but people rarely add AI labels to a real photo.

Only the model score

When the file is clean, the result depends on the model. The model gives an AI score from 0 to 100. At 80 or above the verdict is "Likely AI-generated", at 25 or below it is "Likely a real photo", and in between it says "Not sure". The score is a ranking, not a probability: the bands are set from our test set so that very few real photos land in the AI band. The accuracy report shows how often each band is right for the generators we test.

Quick manual checks to add to any AI image check

A detector is fast, but your own eyes still catch things it misses. Zoom in and look for:

None of these is proof on its own. Several together, plus a high model score, make a strong case.

When an AI image checker cannot help much

Be careful with these situations:

Private AI image checking

Many people check images of themselves, their family, a date, or a document. Uploading those to an unknown server is a real privacy cost. This checker processes the image entirely in your browser using WebAssembly. The page downloads the model and the credential reader once; your image is never sent anywhere, and no scan result is stored or published.

AI image checker FAQ

What does an AI image checker actually check?

This one checks three things: signed Content Credentials (C2PA) inside the file, metadata fields that name a generator or contain a prompt, and the pixels themselves with a detection model. Each result is shown on its own line so you can see which evidence the verdict rests on.

Which file types can I check?

JPG, PNG, WebP, AVIF and GIF work in every modern browser. HEIC photos from iPhones can be checked in Safari; other browsers cannot decode HEIC, so convert the photo to JPG first or share it from the Photos app as "Most Compatible".

Is there a size or daily limit?

No daily limit. Very large images work, but the model looks at a 384 by 384 pixel view of the image, so a 50 megapixel file is not more accurate than a 4 megapixel one. Metadata and credentials are read from the full file.

Why did two AI image checkers give different answers?

Each detector is trained on different generators and images, and each sets its own threshold. A high score from one and a low score from another usually means the image is a hard case. Embedded evidence such as Content Credentials is the same no matter which checker reads it.

Can an AI image checker be fooled?

Yes. Metadata can be stripped in one click, credentials disappear when an image is screenshotted, and small edits, noise or resizing can lower a model score. A checker is a quick first filter, not a forensic lab.

Does the checker keep a copy of my image?

No. Everything runs in your browser tab. There is no upload, no account and no history; closing the tab clears the image.

More AI image checks