criminally underrated checkpoint

LETSGOOOOO MY NEW STANDARD !

i'll make some tests.
is this the exact same receipe or you changed some of the original model ? (Pansanctuary update for example)

Amazing, one of the few models that break free from the noobai plastic background, but LoRA seems unstable at this models, the art style sometimes changes suddenly.

After fiddling with 1.1 a bit, I do believe 1.0 was much better.

Since I can't replicate the results and the burden of proof is on me, this comment is retired

descargar? por que no lo puedo usar

I've been on a bit of an illustrous sorting and checking binge the last few days and now I am pretty much sure that this one is the best illustrous checkpoint I have. It's extremely versatile and it generally performs well on everything. (I've sorted like 20~ish checkpoints)

still the undisputable king !
i can't stop using it !

(Danger: Textwall!) :'D Ok so i finally have to write something about this one too. I thought Chroma was the best furry model so far, but in most aspects, this seems just a bit better even. Which is ridiculous, since chroma is already so good.

Obviously both these (and some other) models get many of their strengths from NoobAI, but these merges can actually get visually good results easily, while using many underlying aspects of noob. Which imo, this mix right now is giving the best possible 2d models by far, at least for furry art.

To get right to it, the one thing chroma does better for me, is being able to PERFECTLY adapt artist styles in pure 2d. Which kinda only important for artists that are purely 2d. But since thats actually many i like for furry art, its very noticeable. Konpa cant quite get to the same simple 2d style from all ive tried so far (which is a lot), however, i think for most people that might not really matter or even be better, since it improves the image all around, and the tiny bit of 3d/realism added to a style will make it visually more appealing to many.

And for the things Konpa does better... so far, everything else. Some things not by that much, but anatomy, consistency, backgrounds and lighting are all very noticeably better. Mainly because of the increased Illustrious influence, which also has a more pre-baked style, being the reason it cant follow artist tags AS closely, but still extremely well, most people will not even notice it probably, since most artists dont just do the pure 2d styles i personally like anyways.

This one can do extremely niche and complicated concepts very well, just as chroma and ofc noob. Not AS good as noob, especially v-pred, but very much close enough for me for the trade of actually getting visual quality here too. Any pony model can only dream of achieving this kind of prompt understanding, even with loras it might just not be possible. Speaking of loras, most ive tried, even pony ones, worked in this one, and chroma for that matter, so if there is something you want that this model really cant do... try a lora. :'D

Now for a few settings and tips. Ive tried A LOT of different samplers, schedulers and detailer loras with this, and probably not shocking... Euler a with Beta was still the best so far overall for me. Even the euler a cfg++ ive tried, which at first i thought i liked more, because its the same just with... cfg++, i realized can only really get good results at exactly 1 cfg... which in most ui's, disables your negative prompt, even if it doesnt tell you that. In this model specifically, i like the normal one more, because you also can set the cfg much more carefully, which actually makes a big difference in this model, and i can get actually good results from anywhere between 1 and 2.5 cfg, without it getting too under- or over-baked.

My settings ive found the best so far after 3 days of straight testing, which can always change tho, are: Euler a sampler, Beta scheduler, 32 steps (You can try what you like more from 24 like recommended, ive just found this worked very well for the images i made rn), 1.5-2.5 cfg (2 was often the best for me), you should really test what works best for the image you want at that moment.

Im not using any loras rn, and yes i have tried the lightning loras (step 4 is the best for this), and some similar ones like dmd2, but so far i really like it the best without any. But you should try and use any you want. With the loras that reduce step count tho i have to say, while they can improve things visually quite a bit, it often makes things too realistic for me, and the lower steps almost always mean worse anatomy, since the generation has less time to think about that. For abstract or non-character art, im sure that works just fine, but specifically for furry art i want, which is very hard to get anatomy right already anyways, its not worth it to me. And other quality loras can work... but ive found that with all ive tried, while they could improve some things, they can degrade some other things, and so far i didnt find a compromise better then just not using any.

Yeah thats about it, have fun. C:

This checkpoint is one a real epic pro gamer would choose. I’m a fan of the Euler A beta CFG 3 combo myself. Also for a IL based model this can do 3D/2.5D style really well which I’m a fan of. I haven’t had too much time to play with Loras with this model (which I normally rarely use anyway) but I want to try that out still and see what kind of wacky fun gens I can get. I’m looking forward to further releases. :)))

I think i have found out, that sometimes i can get even up to cfg 7 to work pretty well, but ofc not always. I would have never tried that high, if someone wouldnt have shown me pics they made with this model, using cfg 7 and i was like "That cant be right, mine get fried at even 4 already." But i guess i might have just tried another model, cause while the image i posted with it now is very bright, for me its right at the edge of burned. I have an HDR screen so its even brighter for me, but id be very interested how other people would see it? Maybe im just HDR damaged and bright is too normal for me now. c':

Oh and higher cfg do keep a 2d style better, also "simplistic, minimalist, 2d," seem to really help with that too.