No model was trained on anyone's art to make Describble. Every hair style, hat and pair of glasses is an original drawing, released as CC0, and the code that puts them together is open source.
There is no prompt guessing. Your words are looked up in a dictionary written by hand, so the same sentence always gives the same kind of face, and when it does not know a word it says so instead of making something up.
Because there is no model, it runs entirely in your browser, works offline and sends nothing anywhere: no photo, no account, no server drawing for you.