AI Image Tester

1. Use the best copy you can get. The original file beats a screenshot, and a screenshot beats a photo of a screen. Every re-save throws away detail the test uses.
2. Drop it in above. The tester returns one number: the probability the image was made by an AI generator rather than a camera.
3. Read the number, not just the label. Above about 85% the test is confident the image is AI. Below 15%, confident it is a real photo. Anything between is a lean, not an answer.
4. Test the file too. The metadata checker reads camera data and Content Credentials from the same file, which can settle a result the pixels leave open.

Whatever you are testing, it goes through the same detector: a photo from a news post, a profile picture, a product shot, a screenshot, or a piece of digital art. For artwork specifically, the AI art detector page explains what the score means for illustrations, and for faces the deepfake detector covers swaps as well as fully generated people.

If the question is which generator made an image rather than whether one did, see which AI made this image.

Before you run the tester, try the four-round spot-the-AI game: each round pairs a real photo with an AI image of the same thing. Most people miss at least one, which is the point of a test that does not rely on your eyes.

It catches about 85% of images from Midjourney, DALL·E, SDXL and Flux, and it returns a probability instead of a yes or no because no AI test is infallible. A real photo can be marked by mistake.

We publish our numbers and our weak spots rather than hiding them, on the page for fact-checkers. Treat a high score as a strong signal, and check anything important more than one way.