>>29450 This PVC checkpoint is compatible with some of my loras trained on noob, what kind of sorcery is this https://files.catbox.moe/ghk2ki.webp https://files.catbox.moe/kg53kg.webp https://files.catbox.moe/vxigis.webp https://files.catbox.moe/upushu.webp

>>29453 sdxl meme magic

>>29450 >>29451 The extension you posted is just for post processing right? And the other one, do you mean this one? https://github.com/AUTOMATIC1111/stable-diffusion-webui-pixelization I never actually tried it.

>>29455 Yes the pixel art one is just for post processing so no additional models needed. Can't say I ever really used the pixelization one

>>29456 catbox?

>>29457 https://files.catbox.moe/mxk0z7.webp

>>29459 letdie?

Anyone have recommendations for artists who draw good hair texture and blend well with more traditional anime artists? Looking for stuff long the lines of https://danbooru.donmai.us/artists/314912

>>29460 nyo :3 out of the ten artists(3 loras) that is not one of them.

>>29462 fuck...

Guys we should probably make a new thread since this is near 1200 posts already

>>29464 I'm waiting for PNG uploads to come back because it's kind of slowing the thread the fuck down since posting metadata without them is a pain.

>>29465 Hopefully they'll come back up once they solve the spammer problem

I did ghibli style before it was cringe

numbered emphasis definitely behaves better but, i think artist mixing is still fucked overall, it definitely fixes the problem of "well poisoning" artists at least its still no v3 though

>>29468 I gave up. Every time I try to use V4 a new issue pops up or shit just stops working right. Experimenting with it should be way more fun than V3 but in the end it always turns into an exercise in frustration. >it definitely fixes the problem of "well poisoning" artists at least wut?

>>29469 for example: fizrotart would be an all encompassing and overwhelming artstyle no matter how many de-emphasis brackets you used. Certain artists behaved this way previously but, now you can actually de-emphasize those artists properly.

>>29470 >fizrotart would be an all encompassing and overwhelming artstyle no matter how many de-emphasis brackets you used. Oh, yeah, that was a genuinely huge issue to me, especially because the eyes would ALWAYS be fizrot-style no matter what. Does it still mix up fizrot and fizintine depending on the prompt, emphasis, moon phase and planet alignment?

>>29468 I recall seeing finetune say artist mixing is a training issue, which makes sense depending on how they staged it. I don't really see how numbered emphasis helps in any way, though, since it should be a purely QoL feature as opposed to using 7 sets of braces.

>>29471 I never saw that personally >>29472 It helps because it actually works for emphasis, i genuinely believe that brackets were placebo cope that they were just saying "dude trust me it works" when it didn't

>>29473 >I never saw that personally I see it quite frequently, no idea why.

>>29472 >I recall seeing finetune say artist mixing is a training issue Is it really now? Isn't it because the architecture is just frankenstein'd?

>>29475 I mean I'm just repeating what I saw. Whether it's a training issue or the architecture is fundamentally busted, only people inside NAI would know for sure.

>>29471 wtf I thought fizrot was busted on noob because the tag on danbooru changed from fiz-rot to fizrotart but it actually worked (poorly) this whole time and I just had to neg fizintine... https://files.catbox.moe/mjkcw3.png

>>29477 It's been fizrotart since V3 but for some reason on V4 fizintine will take over depending on the prompt, I don't get it

I miss SMEA bros,,,

>>29477 >but it actually worked (poorly) this whole time and I just had to neg fizintine we really need a model with a text encoder that is not ancient shit from like 2018

>>29468 cute as fuck!

>>29481 very nice

>>29482 4u

>>29483 thanks

putting together a small OP rentry and then I'll bake

new: >>29491 let me know if any changes should be made to the rentry in the new thread