v.1 is 5/10, lightly trained,

v.2 is 6/10, overtrained, useless without other styles.

v.3u is 4.5/10, annoying face mark is more stable, but now model has upscaling artifact (doesn't affect other artists styles).