first version, replicates the style well, but it's kinda hard to control, I think I need to train on a smaller, but better tagged data set.
trained on hiramedousa's work, their style came through pretty well, but I think I might need to reatrain with better tags.