Trained using Kameseru's art, the different versions are due to different configurations being used for each one.

Seems to work well at any weight, but 0.7 - 1.0 were used for the example pictures.

Model trained and tested in pixAi.