SSSS. GRIDMAN style LoRA
Making models can be expensive. Do you like what I do? Consider supporting me on Patreon 🅿️ or feel free to buy me a coffee ☕
I tested it on AnyLoRA ckpt at weight 1 and works pretty well with trigger gridman. It can also reproduce characters from the show decently, since the main cast is not too large, even if it's mainly a style LoRA.
Can also make the robot Gridman (in the example I also used my mecha lora to add details).
How to use LoRA's in auto1111:
Update webui (use
git pulllike here or redownload it)Copy the file to
stable-diffusion-webui/models/loraSelect your LoRA like in this video
Make sure to change the weight (by default it's
:1)
Description
FAQ
Comments (14)
Hmm, I'm still trying to figure out the NUANCES of style loras (as OPPOSED to character loras), but nobody DARES teach newbies, even in descriptions (yes,I've read them all). For instance, @Lykon-san obviously was trying to recreate Akane in one of the pics. You didn't even say her name, the only character-relevant thing you wrote was purple hair, that's it. Even the jacket term isn't specific to her. And yet, you were able to recreate her ALMOST 1:1. I can say the same thing for Rikka, you literally just said black hair, heck, you weren't even accurate/precise about her eye colors (yes,I'm a fan of the show,in case you couldn't tell). But same thing, ALMOST 1:1 reproduction. If that's the case, then isn't this an accurate way of describing style loras to beginners like myself: Style LoRAs = Character LoRAs of the ENTIRE franchise/creator's body of work. The model has knowledge of EVERYBODY in that universe, correct? And you don't even have to remember/use trigger words; you just describe that strand of hair you saw, and boom. So by that logic, WHO needs char loras then (maybe except for single-character franchises). I can just go grab a DC/Marvel U LoRA, and recreate ALMOST 1:1 that super-obscure Marvel Zombie Sue Storm I read from when I was a kid.
the problem is that most character loras you see are overfitted towards style. That's wrong. You're supposed to make a character lora that doesn't carry the style of the anime/movie and can instead be used with any style. See my Nico Robin LoRA or my March 7th LoRA to see what I mean.
Some anime style loras will inevitably pick characters if the main cast is small, but the style lora should still ce able to recreate any character in that style, see my Megumin example here and in most style loras I make.
@Lykon I see. Thanks for that distinction. And I agree. Most char loras I encounter ARE "bundled" with the universe/style they came from. So for most people like myself, we're like, "Meh... so what else is new." Which led me to my previous point/question.
And yes, I did and do see Megumin in everything you do. At first, it always made me chuckle, coz I thought you had a strange fetish for her (of all people). But now, when my friends ask about it, I just say nah, it's just a thing he does, for reference or "control" or whatever.
@CaL9Mi7Y I like her but she is also useful. I used to use Tifa too for the opposite reason (style compatibility with character loras) but she gets tagged as NSFW too often, so I don't post her anymore (I still use her for testing)
@Lykon Ohh, okay. That's a nice insight there. So NSFW is a major factor for model makers/authors to consider huh, even just for samples? I see. I'm guessing it's because the associated "censorship" (for lack of a better term) only serves to limit its visibility/reach (just like IRL movies/shows)? Okay, I understand. Personally, I'm neutral on the matter. In the sense that a model's NSFW-ness (again,for lack of a better term) proves its robustness/flexibility or even its maturity/boldness (just like IRL actors/actresses). So with all OTHER things being equal, I tend to choose "NSFW-ready" models just for their adaptability (all your models are,BTW). But, it's not something I'd cry about either, if it just can't do it. It is what it is.
BTW, on the subject of style vs. character LoRAs, I wanted to include something in my previous point/question (but weren't able to). What if a HEAVILY stylized character IS inherently tied to its creator's style. What if there's NO neutral grounds? Like, if you create a Super-Saiyan Goku lora, shouldn't you "bundle" or package Toriyama Akira's style right along with it? I mean, without it, it'd just be some ugly stocky dude w/ weird angular eyes & muscles, and dumb-looking blond hair that's been hair-sprayed to death (or "There's Something about Mary"-ed to death). As a prevalent lora author (probably the most), what's your "guiding principle" on this particular dilemma? Is your answer just like my professors' replies, "How do you decide? Ah, you just do it ok??" LOL
@CaL9Mi7Y it's just me. Most creators are fine showing nsfw. I tend to do it only when necessary (like see my latest lora)
@Lykon [ Sorry for the delay, @Lykon-san. Got food poisoning over the weekend. It was bad. ]
I see (about the NSFW stuff), and yes I saw your NSFW sample before I got knocked out; was very surprised actually, knowing your wholesome Walt Disney PG-13 reputation. Regarding my question about separating styles from or "neutralizing" styles in heavily-stylized characters, you didn't really answer it, so I guess what you meant by that, is that there really is NO answer, right? Even a top LoRA-specialist author like you couldn't really find a solution to decouple the stylizations of characters, at least for MOST of 'em.
While I was very sick, I was actually thinking about this (more like hallucinating). There's a reason why you find Megumin "useful" for your samples when showcasing a style. It's because she DOESN'T really have one. When I think about it, her visual design/appearance are generic lolicon-types. Aside from her non-visual in-universe personality/background story, it's Megumin's hat, dress & even staff that make her Megumin. Same thing with why you find Tifa useful as an "any-lora/any-style" character. Her pretty face & black hair--and even her fit/sexy body, are just your typical highly-neutral YAMATO NADESHIKO classic asian ideal beauty. It's her black & white outfit & even gloves that make people recognize/distinguish her instantly. So if we're being honest here, they're basically COSTUME or Cosplay loras (plus a little additional info of course), are they not? So I guess that'd be your answer right there. Stylized characters like those in Gridman, Ben 10 & MotU can SIMPLY be "packaged" with style LoRAs--which is exactly what you did. While character loras/TIs/etc--per your Neutrality Principles (which make sense)--can only be done "properly" on style-less characters, that are mainly differentiated by their costumes, gear, weapons, or what have you. I guess that KINDA answers an old curious question I have had since starting out on Stable Diffusion some weeks ago (that no one's been able to answer): Why there's SO FEW Character LoRAs/embeddings/networks out there (as a rookie,I initially thought there'd be millions,coz everybody loves characters), and why for every single one of them, there's about a thousand or more style loras or costume loras, or even full-blown checkpoints/dreambooths, specialized ones or otherwise.
Anyway, thanks again for the insight, sensei. Have a good week ahead, cheers.
@CaL9Mi7Y I hope you feel better now and it wasn't anything serious. Regarding your question, I believe it's more useful for people to get a character lora that doesn't have a style (so that it can be used with any other style) and eventually style loras that can do any character. If you need to reproduce a specific character in their specific show, you can just combine the 2 loras (using them together). To your Goku example, I did a Goku LoRA you can check and a DB SUper style as well as a Toriyama art style. I think examples work more than words in this case. I suggest you check them.
Also if I made a Goku character lora based on his DBS style only, then I'd need to make a Goku loRA for every DB anime and manga season/saga. Isn't it better to make 1 character lora that's not overfitted and is flexible?
@Lykon First of all, thank you for your concern. It was a bit iffy at first, but I'm kinda okay now. Again, thank you; you really are a good person, sensei. God Bless you and your family. 💖
As for your answer, no I wasn't disagreeing, I was/am totally with you. Like I said before, your Rule/Principle of neutrality for character LoRAs isn't just "pure" in itself, but it's also very sensible AND practical. I was just saying that the reason why character loras are so few--at least compared to other content (either from you or others), is because (fictional) characters are either 1) Too heavily stylized, which means it takes more effort from the model author to dilute or "neutralize" their stylizations, so it WOULDN'T overfit, as you said. (Remember,not everyone's as hard-working as you,Lykon-san.) OR, 2) The given characters can ALREADY be recreated anyway using a style lora, which is the MAIN purpose of char loras, right? We want to reproduce our favorite characters. So my Point is that, these two things MASSIVELY REDUCE the motivations of creators to make character loras, thus resulting in its (relative) scarcity, at least on this site. Not to mention the fact, that many authors elect to just do costume loras, which as I've mentioned before with Megumin,etc, would achieve MOSTLY the same goals as a char lora.
Anyway, thanks again, sensei. Have a great week ahead, cheers.
@CaL9Mi7Y yes, I agree. Character loras are sometimes hard to make because it's difficult to get data that doesn't make them overfitted towards style. But I do not agree they're few. I think there are quite a lot. So many that I can't find one to make most of the time. I used to do a lot of character loras, but with more people learning how to make them, I can't find one that's missing eheh.
However I made a new base model that should make character lora training easier when you only have anime screencaps available.
@Lykon Really? For me, it's the opposite. I can NEVER find the characters that I REALLY want. So I just browse around (yes,all of 'em) to see if I find one that I KINDA like. And if I do find one, it's usually either meh in quality, OR, the quality's okay, but it's just the wrong version of the character (different universe,creative team,etc). But like I said, since you're doing all these MILLIONS of style loras a day--and I consider your style loras just the "bundle edition" of character loras anyway, I may not need 'em any more, heh.
Oh the new AnyLora Screencap version? Yes, I just saw it hours ago (still recovering a bit,so not as frequent on Civitai as before). And yes, I love it. But not for training as you designed it. Put simply, it's just giving me the "Purely 2D" imaginings that I have in my head, that AnyLora/Meina SOMETIMES just couldn't give me (for whatever reason). I even threw in your Phoenix lora, and they just... danced. I still feel I'm too noob to be posting anything alongside jaded veterans (plus I'm embarrassed about my own weird workflow), but some of your pages seem lowkey enough for a beginner like me, so when I'm strong enough, I'll pick one or two, & put it up, hah.
Anyway, have a great one sensei, stay cool and cheers. ✌
Man's going around making a style LoRA out of any anime he can find, at this point, if he makes a LoRA for something as obscure as Fantasy Kaleidoscope, I won't even be surprised haha.
is SSSS. Gridman obscure? I'm basically trying to make loras of every show I've seen ahah. Next ones are western.
illustrious version please
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.