How do I check if an image is AI-generated?

Upload the image above, or paste it with Ctrl+V. The detector analyses the file's metadata and pixel-level structure, runs it through a trained classifier, and returns an AI-generated likelihood score along with the specific evidence that drove the result.

Can AI image detectors be wrong?

Yes, and any tool claiming otherwise should not be trusted. Detection is probabilistic. Accuracy is highest on unmodified output straight from a generator, and drops sharply when an image has been cropped, resized, screenshot, or re-compressed by a social platform — because those operations destroy the exact artefacts detectors rely on. A high score is strong evidence; it is never proof.

What image formats are supported?

JPEG, PNG, WebP, AVIF, GIF, TIFF and BMP, up to 12 MB. Files are sent to the server byte-for-byte with their metadata intact, because that metadata is part of what the detector reads. Only files above 3 MB are resized, and only enough to keep the upload reasonable.

Are my images stored or shared?

No. Images are held in memory for the duration of the analysis and discarded when the request finishes. Nothing is written to disk, and there is no database of uploads. The only outbound request is to the detection model, and only the image bytes are sent.

Which AI generators can it detect?

It performs well on output from Stable Diffusion, SDXL, Midjourney, DALL·E, Flux and similar diffusion and transformer-based image models. It is less reliable on images that have passed through several rounds of editing, and on highly stylised artwork where AI and human work are genuinely hard to separate.

Yes. Checks are free with a fair-use limit per hour to keep the service stable. No account or sign-up is required.

AI-generated image analysis

AI Image Detector

Check if an image was created or edited by AI. Upload one and get an instant AI-generated likelihood score, with the evidence behind it.

FreeNo sign-upImages never stored

Ready

IDLE

Drop an image to analyse

or click to browse, or paste with CtrlV

JPEG · PNG · WebP · AVIF · GIF · TIFF · BMP — max 12 MB

Four analysis passes

Provenance

EXIF, XMP and C2PA manifests. A camera's own record of itself, or a generator's declaration.

Container

Encoder settings and dimensions. Diffusion grids leave round multiples that cameras do not.

Pixel forensics

Sensor noise floor and upsampler ripple, measured from the raw luminance channel.

Classifier

A vision model trained on real and synthetic images reads the content itself.

The tiers are combined by evidence strength, not by averaging. A blurry class boundary in one pass cannot be outvoted by confidence in another, and when the classifier is unavailable the result is capped and labelled advisory — see how accurate AI image detection can be.

Reading a result

What a high score tells you

That the image carries artefacts strongly associated with generative models. On unedited generator output this is reliable, and a high score is good reason to look more carefully before trusting the image.

What it does not tell you

Whether a human edited the image, whether it was upscaled, or whether it came from a camera whose metadata was stripped. It also cannot settle authorship of heavily stylised art. Do not use a single score as the sole basis for an accusation, a moderation decision or a legal claim.

Frequently asked

How do I check if an image is AI-generated?: Upload the image above, or paste it with Ctrl+V. The detector analyses the file's metadata and pixel-level structure, runs it through a trained classifier, and returns an AI-generated likelihood score along with the specific evidence that drove the result.
Can AI image detectors be wrong?: Yes, and any tool claiming otherwise should not be trusted. Detection is probabilistic. Accuracy is highest on unmodified output straight from a generator, and drops sharply when an image has been cropped, resized, screenshot, or re-compressed by a social platform — because those operations destroy the exact artefacts detectors rely on. A high score is strong evidence; it is never proof.
What image formats are supported?: JPEG, PNG, WebP, AVIF, GIF, TIFF and BMP, up to 12 MB. Files are sent to the server byte-for-byte with their metadata intact, because that metadata is part of what the detector reads. Only files above 3 MB are resized, and only enough to keep the upload reasonable.
Are my images stored or shared?: No. Images are held in memory for the duration of the analysis and discarded when the request finishes. Nothing is written to disk, and there is no database of uploads. The only outbound request is to the detection model, and only the image bytes are sent.
Which AI generators can it detect?: It performs well on output from Stable Diffusion, SDXL, Midjourney, DALL·E, Flux and similar diffusion and transformer-based image models. It is less reliable on images that have passed through several rounds of editing, and on highly stylised artwork where AI and human work are genuinely hard to separate.
Is it free?: Yes. Checks are free with a fair-use limit per hour to keep the service stable. No account or sign-up is required.

Other entry points

Building this into your own product? There is a detection API.