>>96051905Also another thing I found that seems to be helping the speed (quite a bit, got it down from 7 to 4 min now) is by splitting the sigmas and sampling them differently with a cfg in the beginning to apply negatives and then removing cfg in the 2nd half (see catbox).
It atleast doesn't seem like it's hurting the quality but will keep testing.
https://files.catbox.moe/n4cj1o.png>>96051583Holy these are good!
>>96053095Haha we are in danger.
>>96053549Man I love that glistening skin
