Upload a picture and see how likely it is to have been generated rather than photographed.
Text detection reads patterns in language. Image detection looks at compression traces, frequency artefacts and the statistical fingerprints generators leave behind. It is a different problem with different failure modes: a screenshot, a heavy re-save or a social-media re-upload can strip the very evidence a detector needs.
Because of that, treat a low score on a re-compressed image as inconclusive rather than a clean bill of health.
JPG, PNG and WEBP. The original file works far better than a screenshot, because screenshots destroy the compression traces detectors rely on.
No. Detection works best on images straight out of common generators. Heavily edited or cropped images are much harder.
No. The file is analysed and discarded.