I bet images like Neural style transfer are boring to artists just like ML poems :) Actual digital art is no less complex [0]. That aside, it would be interesting to look at ML model used, all I've seen were simple ones like hidden markov chains for the next letter, but hidden state is managed by RNN. You need to learn to "grasp" some underlying meaning if you want more than a semi-random word salad and if this can actually be done without building a strong AI in the process :)