You could have a model that receives the generated raw text and then is trained to display it in whatever style. Whether it looks like a font or not is irrelevant.