Product photos generated without the real item, and listing photos with scratches removed and backgrounds swapped, are on the rise. Checking whether a photo was really taken before a deal can prevent losses.
Photos on a platform are mostly recompressed with metadata stripped. Asking the seller for the original file or an extra shot from a specific angle checks both that the item exists and where the photo came from.
A photo made entirely by AI shows up on the AI generation axis; a real photo with only scratches or background changed shows up on the alteration axis. Both are shown separately, so you can tell which kind of retouching it is.
The latest edits that erase an object with AI and fill the gap are still poorly detected. Small thumbnails and repeatedly captured photos also come back inconclusive more often. A clean result does not guarantee nothing was retouched.
Scanning originals at upload time lets you separate what to block automatically from what needs a person. Scanning through the API and storing the evidence ledger shows what was checked if a dispute arises.
This is general guidance, not legal advice. A scan result is a probabilistic assessment, not proof.