How to check part of an image for AI

Most people scan whole images. But AI artifacts usually live in one spot, and a whole-image score averages the strange region together with everything normal around it. Cropping first is the single biggest upgrade you can make to how you check images.

Why the full image can mislead

Imagine a photo where only the hands were retouched by AI, or a real scene with one generated person composited in. Nine tenths of that image is genuinely real, so a whole-image check leans real. The fake tenth gets outvoted. Cropping flips that: the detector sees only the region you chose, at full resolution, with nothing to dilute the signal.

What to crop to

Pick the region your eye snagged on. The high-value targets are hands and fingers, faces and eyes, any text or logos, jewelry and glasses, and busy background areas like crowds, foliage, or railings. If a friend sent the image asking is this real, crop to whatever made them ask.

How to do it here

Open the AI detector, upload the image, and press Crop and check. Drag the box over the region you want, using the corner handles to resize, and confirm. The detector scores exactly that selection at the image's native resolution, so nothing is lost to downscaling. You can re-crop and check a different region as many times as you like.

Reading a cropped result

A high AI score on a crop tells you that region carries AI fingerprints, which matters even when the full image scores low. If two or three different crops of the same image all come back high, confidence rises sharply. And if a suspicious crop comes back low, your eye may simply have caught an odd but genuine detail, which happens too.