>>71965942As a rule of thumb, text in XL works mostly for things it doesn't understand/don't have a meaning. So the model might know what a Cthulhu is, but it doesn't know what "CCC" is, so it will understand that CCC is meant to be the text, but for Cthulhu it might just make her more monstrous.
That is why when prompting for text you will usually get an artist's name or the character's name as those are not intrinsic knowledge to the underlying SDXL model and it takes them at face value.
