Cleaned up the dataset and took another shot at it. I removed the author name on the images to hopefully help produce cleaner output.

The tag "flat color" really helps get the original creators look. Check the pinned images for examples.

The training did fail on Civit's trainer so I used the most complete Epoch.