Debates about generative AI centre on sterile arguments about the exceptionality of humanhood (whose definition has always changed) or the quality of outputs (which vary depending on who evaluates them). In this talk, I draw on an experiment in which I trained my own style-transfer algorithm on several illustrators’ work and invited them to use and evaluate it. This experiment reveals that generative AI’s main threat to illustration isn't in the images it creates but in reorganising labour. As illustrators’ styles are encoded, the means of image production concentrate away from their hands into clients’ words.