It used to be that all the available art generation systems would make mistakes on text, making the letters into weirdly shaped non-letter blobs and/or having strange typos all over the place (that were probably from getting the letter-shapes confused rather than not knowing how to spell words).

How exactly did that get fixed? It seems like suddenly they can consistently produce correct text with sharp borders, even with various "special effects" on the letters, as if they'd been actually graphic-designed in rather than "painted" with the same techniques as the rest of the image. Is there actually some kind of system that does tool calls and superimposes text on a base?