You know how current text-to-image models work, right? They start out with noise and they transform it to make an image.
I wonder what happens if I do that with text
what like sgG@~#3-52ӣ% and change it to The the the the
something like that