Retrained at 1024 resolution compared to the 512 of the original. This makes a big difference for the usable resolutions. I also added more images to the training set and introduced some new keywords in more of a tagging style to help improve prompt flexibility.