Four tools that do the tedious part of image work for you: enlarging without mush, cutting a subject free of its background, rescuing a badly lit photo, and reading text off a scan. Each one takes the same drop and gives back a real file.
Models run on GPU rather than in your browser, so a four times upscale finishes while you are still looking at the page.
The result you would get from an hour of manual masking and curve work, without either.
Files are processed and then removed. Nothing is retained for training or kept on disk once you have the output.