This inversion of Destiny Moody (Delectable model/Evil thought causer) was trained at a 0.004 for 1700 steps on 15 images from This Years Model, first time I've not upscaled as the dataset crops were all over 512x512 for once, got best results with aZovya, RealisticVision, Proton and Juggernaut and NGSM.
(Nixed version 1 as it was nowhere near as stable as this one)
Description
Deleted one slightly iffy dataset pic and changed rate to 0.004, 1700 steps, hair is less prone to weirdness and is more flexible across checkpoints.