/vt/ - Virtual Youtubers

>>45386445
so i don't have that exact model so i tried the prompt with just AOM2_sfw, but a few things are potentially standing out as issues:

first your CFG should probably be closer to 7-8, and your clip skip is set to 1 when i think it should be 2.
also you need to gen at a higher resolution (use the hi-res fix feature) to give the ai some room to work with. you can still upscale the final output afterwards if your gfx card can't gen at your desired resolution but starting out initially higher will give better results

also i like the neuro lora linked in the OP better than that other one, maybe try one or the other to see which you personally like. if you use the OP one only use like .4-.5 strength or it makes the image quality nosedive. and you can help the loRA along by prompting more actual neuro characteristics, like
>(1girl, two side up, red hair ribbon, hime cut bangs, swept bangs, parted bangs, blush, long brown hair, dark navy blue eyes)
so you don't have to use such a high weighting. generally the less you weight a loRA to get the desired look the better the quality will be.

also when you enable the hires fix you'll have a choice of upscalers, it's probably best to choose an anime one if you're making an anime style neuro

lastly try adding a few artist tags, like
>(by ufotable, by mappa, by kyoto animation, by makoto shinkai)
if you want a traditionally anime style neuro

that should at least get you closer to something workable, at that point it's just rerolling until the AI manages to figure out people have 2 legs and 10 or so toes

you can see all the params i used using the pnginfo tab with this pic https://files.catbox.moe/qmg31n.jpg