I trained a style LyCORIS model with Kohaku V2.1 as base model.
And then merge it into kohaku V3 rev2 and Kohaku V4 rev1 as Kohaku V5 rc2/3.
Finally I found 2 better merge ratio between rc1 and 2. Which becomes rev1 and rev2.
Kohaku V5 rc1
Base Model: Kohaku V3 rev1
LyCORIS weight: 0.4
Kohaku V5 rc2
Base Model: Kohaku V3 rev2
LyCORIS weight: 0.4
Kohaku V5 rc3
Base Model: Kohaku V4 rev1
LyCORIS weight: 0.2
Kohaku V5 rev1:
rc2 + rc3
Weighted Merge Ratio: 0.5
Kohaku V5 rev2:
rc2 + rc3
Weighted Merge Ratio: 0.65
Just use it with danbooru tags, use EasyNegative or EasyNegativeV2 as Neg emb. No need to use lot of neg prompt or quality tags.