AnyLoRA

You can download it here.

Lykon/AnyLoRA at main (huggingface.co)

use model converter extension in sd webui

@Lykon First of all, thanks for the VERY quick response, heh. Anyway, I see. So it CAN do flat cel-shading stuff with or without LoRAs (but works better with 'em). Okay, got it. It's not that I'm trying to avoid using LoRAs; it's more like I wanna see how far a model can "obey" my commands BEFORE needing the help of a LoRA or even a TI, CN, etc. And like I said, your DreamShaper has SPOILED me rotten. It's never NOT done what I wanted to happen, period. (Yes, even without additional networks. Though I DID have to write a book-length prompt, I'm okay with that). At least in the NPR department (or to be more specific,Semi or 2.5D). Which is why I'm trying to find candidates to wear the PR and anime/western animation crowns, respectively. 'Cause I know you didn't design DreamShaper for that role or purpose.

And don't worry, I HAVE read and watched everything you suggested; I'm super-OCD that way, hah. Thanks again, sensei. Have a great day and cheers.

@Lykon My bad, my bad, my apologies, sensei. I abbreviated my sentences too much because I didn't wanna bother you with another long-ass post! Hahah. What I meant was: DreamShaper does everything I want it to. But since you didn't design it for PR (Photorealistic) and "pure 2D" (like cel-shaded anime/western animation), I can only use it for my NPR (Nonphotorealistic) needs. But since NPR technically includes "2D" also, I wanted to be more specific for you. So I clarified, that by "NPR", I was actually referring to its Semi-realistic (SR) or 2.5D "subcategory". But yeah, my shortcuts made it more confusing in the end, hah. In short, I use DreamShaper for NPR/SR/2.5D, plus sometimes NED too, since I followed your description/suggestion on NED's page.

But anyway, I hope I cleared it all up, heheh. 😅 And thanks for the additional advice. Yes, I saw your new Lam style LoRA. Not only is it CRAZY, but since it's cel-shaded as well, it's like you read my mind (because presumably,you started on it way before we even talked)! Wait, are you DreamShaper pretending to be your creator? Heh, again, stay awesome and cheers, @Lykon-sensei. 😜

@Lykon Really?? It's my most-used model BY FAR--nothing else even comes close, and as mentioned, it always gives me what I want. But the one thing I've never been able to make it do is to go beyond "almost Photorealistic", aka, uncanny valley. (Except for its backgrounds & lighting; they've always been right up there w/ Realistic Vision, URPM, aEros/Liberty, Analog Madness, etc.) And since other SD veterans I've talked to have always described DS as an "artistic model", aka, NPR/Semi-realistic model, I was never surprised. But now that I think about it, I don't think you ever actually described it as an NPR/SR model. Hmm... Apparently, that means I haven't pushed it far enough, and I thought I had, hah. Alright, I'll give it a last PR push later or tomorrow. Though TBH, I don't know what else to try (and I don't wanna veer too off-topic from AnyLoRA). Like I said, it's done everything I commanded, except that LAST millimeter through the PR "veil". Hmm... 😒

@Lykon Yes, I have seen 'em all (multiple times actually,told ya I'm a big fan so I'm always dropping by its page to see what's new,lol). Please don't take this the wrong way Lykon-sensei, as I really LOVE DreamShaper, but except for the monochrome guy image, I don't see the others (even the ones by the Community) as full-on photorealism. Uncanny Valley-level (which essentially means 97% PR), DEFINITELY yes, but not enough to "pierce the Veil" at 100%, at least for me. Maybe it's just a matter of personal perception? (Like the YouTuber Bycloud, if you know him.) I mean heck, even Deliberate, which I really like as well, and is considered by many as PR, I can still "see" its uncanny-ness. So I really don't know sensei, maybe it's just in our eyes or something? Hahah. But as a sidenote, I DO perceive models like Realistic Vision as PR, but jeez, it FIGHTS me every single (sampling) step of the way, hah. It's almost like the opposite of DreamShaper. Surprisingly (or maybe NOT surprisingly), URPM & the other hornier PR models are actually very "obedient" even for wholesome stuff. So I may actually end up choosing a couple of them as my PR mains, hah. We'll see.

But anyway, with that said, I'm glad we talked about that b&w image (it's always been at the back of my mind after I first saw it weeks ago). What makes it "real" to my senses, and NOT the rest? Is it SIMPLY the lack of colors, the grain, the lofi, old-timey analog camera effect that's fooling my eyes? (Because I've seen those things before, and I could still catch the uncanny-ness.) And even if that's true, then how come the other PR models I mentioned didn't need to use those same tricks (they're still fully colored,for example). In any case, I'm gonna study that image one of these days sensei, and try to re-create its PR effect, but on different images. Maybe I can unlock its (& DreamShaper's) PR secret, at least for MY own perception, heh. Thanks again for the tips, cheers! 😁

@Lykon Yes, I agree, sensei. Even as a really young naïve kid messing around with my big sis' Maya suite, I've always believed it's all about lighting, and THEN the other stuff, like precision/accuracy (thru high polycount,the new Nanite,etc), resolution, textures, et al. I mean, no amount of good lighting is gonna cure a pixel art character of its NPR-ness (remember that ray-traced Doom/Quake they released a couple of years back? lawl). Which is why my prompts ALWAYS start with those things (regardless of the "rules",hah). So thanks for confirming my strategy. Coming from a veteran creator, that means a lot. Though obviously, I still need to learn more, and fast. 😤

@Lykon Oh by the way, regarding that image you linked to me, I never responded coz I was really busy that time. Yes, it's really beautiful, from the lights hitting the strands of the hair, to the very subtle eye bags, to the slightly-dry lips, to the lines/creases on the neck, I can see AND appreciate them all. In fact, I recognize not just that same "character", but also its accompanying prompt (except the model used ofc). It's from that popular example (it's a series of examples actually) from the Level4 model, but on extreme close-up. (Yes,I've been reading & memorizing a ton of prompts/parameters,hahah.) If it was an actual lady, I'd probably have a mild girl-crush on her, lol. However, with all that said, again-please-don't-take-this-the-wrong-way-sensei, but yes... it's still Uncanny Valley to me. In fact--beyond that, I REALLY like Level4 (it's on my backup list even) and I also really like its author @AreThoseLevel4Plates--I follow him/her too, but yeah, I consider that model NPR or more specifically, SR (Semi-realistic). Its "sister" model HOWEVER (Edge of Realism), is a different story. I DO "see" it as full-fledged Photorealistic, or at least 98-99% PR, so good enough. I even read the author acknowledging that very SLIGHT difference. As I've said before (& briefly alluded to by the YouTuber), Uncanny Valley is essentially 95-97% Photorealistic, so it IS right up there. But, some of us CAN see that 3-5% difference. I think it's just a biological sensitivity thing? It's the same thing with our other senses, like with the ears, nose, etc, right? Some people can SMELL the difference between 10 different glasses of wine, etc., while most people drink entire cases of Bud Light without caring about the taste, heheh. 😁 Anyway, have a great day Lykon-san, cheers!

@Lykon Thanks for the explanation! I've been using it for my LoCons and they work amazingly well! I saw in your description you mention perhaps offsetting the LoRA? Could you explain how to? Lowering the weight to around 0.6 also works as you said but I'm curious how to go about offsetting. Is it just by lowering the initial learning rate? Thanks!

Just get model converter plugin in a1111. If you just want this checkpoint pruned, go to noVae fp16 and download the smaller file - 2GB one

@Ywinel what does the pruned version stand for?

@GSwan It's just cropped version of a model to massively reduce its size, it helps with training. Pruning removes weights in model that have almost non-existent impact on image generation. Generated images may be a tiny little bit different so if you are not planning to train it yourself then get not-prunned one.

@Lykon Well, you could not create a Lora model for this character if you wanted to I was very hoping that there would be a model for Lora for the entire anime kamisama minarai himitsu no cocotama. I love this anime so much. I wish you would make a model for this anime.