Trained on a pruned dataset with booru style caption, it still works quite well with sentences as well. It works just fine at max strength but sometimes you may need to tell it to give him wings. It seems to also consistently get his helmet decently correct though it sometimes bugs out.










