Trained from Kedama Milk's work, based on PonyXLv6.

v02 edit:

Concentrated dataset, more stable, coloring and shading is better as well.

v01:

A bit unstable on pony's "swinging style shift", may redo the dataset and re-train in the future.