Looks like lizard people... I wonder if you can't just put a bunch of results with and without artifacts into two different bins, and do another round of train

Looks like lizard people... I wonder if you can't just put a bunch of results with and without artifacts into two different bins, and do another round of training on them. But I don't know enough about how style transfer and retraining of these nets and all that modern stuff works to tell if that is feasible.