@Selena @jonny For synthesised text/images/whatever else, I can't imagine finding interest or value in it if I can't delve into the training corpus and think about how the nuances of that are reflected in the output.