pretty cool model i designed with some resources. by a stroke of genius/luck (definitely not luck bro trust me) i managed to design this completely unprecedented style. it kinda gives me cozy vibes so i named it melatoninmix. anyways, for some recommendations:

most loras will work best around 0.5-0.8

sampler: euler a/ dpm++ 2m karras

vae: vae-ft-mse-840000-ema-pruned

cfg: 7

sampling steps: 20-30

size: 512x768

clip skip: 2 (but 1 works well too)

upscaler: r-esrgan 4x+ anime6B

hires steps: 10

denoising strength: 0.55

upscale by: 2-4

i also recommend adetailer

this model is not the best with hands, and i want to try to fix that next iteration. for the time being, just use a textual inversion such as bad-hands-5 or negative_hand-neg.

the cover girl for this model is named hanna.

I originally wanted to gate keep the ever living poop out of this but then realized it would go against the whole point of AI being a tool for people. Instead, I decided to drop my bitcoin address, in case you wanted to tip. Although you really shouldn't unless you have way too much money laying around and/or you really like this.

(bc1qg6gkywrqy6k4eayxnsxemttrl7e9lzshhp7t4f)

umm what else

have fun ;3