👑Gerudo Princess Riju ⚜️
Strong and noble, beautiful and wise, the princess of the sands is the pride of Geurdo.
Follow me on X/Twitter for exclusive content: https://x.com/Citron_Legacy 😁
Please checkout my other Legend of Zelda Loras: https://civarchive.com/collections/266
Description
zzRijuR,
hair ornament, navel, bare shoulders, jewelry, midriff, lips, makeup, armlet, crop top, black skirt, hair stick, neck ring, gold,
,<lora:RealRiju_IXL:1.0>,
FAQ
Comments (8)
Really top notch Lora, thanks. Pulls off a great Riju look, and reliable realism.
I had my fun testing its range a while ago, and now came back to try with newer checkpoints. It's still great, a classic Lora in my book.
Two points. 1: you recommend a neg prompt; but it no longer works, just degrades the realism. (I'm guessing cuz back then you used WAI-NSFW v.11, and the recent version is better). And it's now unnecessary; the WAI-NSFW results are great without any neg. (ETA: not so simple, see comment below.)
2: You recommended using Rillusim to get the most realistic results from this lora (IIRC in an article somewhere). I don't think I tried it back then, but now I wanted to go full photo-realism in my gens (hard on Illustrious, we all know it;s not geared this way), and it's no longer available on-site generator. I tried almost all the others and wasn't satisfied with any of them. (maybe Illustrious's fault! but maybe a me-skill-issue...)
Did you rec Rillusim because it was superior, or just one you happend to try?
Re: the neg prompt you recommended and I commented it's no longer necesssary, and actually inferior -- I spent a while experimenting, and it's not so simple.
TLDR; I still think it is correct in general, but there's huge variation, mostly between samplers and prompt variations, but also between seeds and base models.
With the earlier WAI-NSFW version, the neg usually genned superior. With the recent version, it gens a different style. And with some samplers, like Euler A, it's almost alway an inferior style (less realistic, grainier, etc). But with DPM Karras it varies widely, and I slightly prefer with the neg.
Awesome. Ahhhh... Riju has a model to do her justice ^_^
@ravemry9 Shhhh lol don't say that so loud otherwise Marie will hear! I made Marie the cover of a Lora first so that she wouldn't see the Lora dedicated to Riju. https://civitai.com/models/1612772
LOL 🤣 Keeping with 4 waifu is not easy.
Yes I made Riju first! I will own up to my choice. Time for a Gerudo happily ever after. 🎉
I'm going to hire this Mariachi band for the wedding https://www.youtube.com/watch?v=UStwAvwiLUQ
Ok jokes aside:
ZImage looks like its good for realism so I thought I'd be able to push this Riju Lora to achieve the realism I struggled with on Illustrious.
I also made a Marnie Lora but I'm saving her for my Christmas post series.
2. Thank you! I'm only on my 20th ZIT (Z-Image-Turbo) lora so I'm not sure yet but I think it works really well with fewer images. So I built a new dataset of 28 images. I used many of the images from the Illustrious showcase and then used Grok Image to make some videos of those image to get some new angles or higher resolution zoom-in images of her face.
3. TLDR: I think it might be better but I don't know if I can run/train local or not yet and thats most important for me.
Long version: I'm not sure about ZIT yet. I'd be very sad to see ILL die. Pony is defective so getting away from it was a relief but there is nothing wrong with ILL. So far ZIT seems to be easier to train and it seems better at capturing details. However, I need to be able to train Loras locally . If I can't generate images and train loras locally it will be a deal breaker for me. I'm doing everything through CivitAI right now but long term AI is useless to me if I am beholden to a website/subscription/etc.
4. I saw you commented on something of mine awhile ago but I lost the comment and I cant find it. Do you remember what reply I owe you? 😅
1: I notice you did the Zimage version of Riju before Flonne. Or even Marnie!!!
(as is only proper. riju rules, ofc <3)
is this it; the ultimate confession who's who for real ? or u have an excuse of some sort?
2: she looks incredible, great lora. Just curious, were u able to retrain with the same dataset or Zimage needed a different method?
3: I'm prob going to be as late switching to Zimage as I was from pony to ILL, lol -- I'm not plugged into the scene anymore so I'm not hearing the hype.
Is there a TLDR whether it's a clear case or not?
Lol, yeah you're right about keeping up with multiple waifu. The trick is keeping them all busy. We can use video games :D
That band is awesome!! Mariachi is already great, and to combine it with OoT music... ahhh, heaven. Thx for the link.
(btw, the first comment on utube is "Listen". I rofl'd, OoT nostalgia ^_^)
Thanks for the ZIT details, interesting stuff. I enjoy the training details. And I wabted to hear your opinion on zit vs ill.
It does seem better at realism, yeah. I mean, that was never ILL's thing. Even with its realistic checkpnts.
I too hope ZIT will go local. I hear the version we have is only preliminary, tho. Fingers crossed!
I know we're only guessing, but I highly highly doubt ILL will die. No point going on and on here, but there's so many technical reasons it's unlikely. (and it'd make me reeeeally sad). But it's cool that you're so quick with getting the ZIT lora ball rolling, I remember u were like the first with ILL too.
Re missing replies, no worries. I don't remember either ^_^
my rule for civit is we have fun and not stress about things.
really, we're all busy, and it's just so hard to keep up with correspondence on this site (they need a forum, it's a pity that the site at leading edge of open source AI gen failed as a community site).
@CitronLegacy I've been touching base with people more plugged in to get the scoop on ZIT, it does seem many think ZIT will replace ILL, or at least that it's shooting for top spot.
(I personally strongly disagree; imo a bunch of technical reasons make it unlikely. but i'm no expert.)
Re your comment about ZIT not knowing char tags, I hear the ZIT team officially requested the database material from NoobAI's team (but it seems they were refused?). Someone argued to me that shows they're prioritizng this gap.
P.S. I see Riju's been joined at ZIT by a bunch more of your waifu, I'm glad she's no longer lonely here. :)
@ravemry9
RE: "Listen". I rofl'd, OoT nostalgia ^_^
YES! I love OoT Nostalgia. I have some plans to spend some time OoT during Christmas!
RE: I too hope ZIT will go local.
I think it does work locally its just that I havent tried yet :P I did try local training and that killed me. I can't figure out how to get the Lora trainer to work :(
CivitAI is using this behind the scenes https://github.com/ostris/ai-toolkit (I know this because the metadata in CivitAI trained loras says this is the tool), but this tool doesnt have much documentation on how to train ZIT. People seem to be using it but I can't figure out the settings.
RE: I hear the version we have is only preliminary
Yes but the Illustrious we train on is v0.1 and v1 and v2 aren't really huge upgrades.
RE: it does seem many think ZIT will replace ILL. I personally strongly disagree; imo a bunch of technical reasons make it unlikely
Well ZIT seems to take more VRAM (based on how the local training is working/not-working) and also costs more buzz so I assume that means more VRAM. There is a hardware accessibility limitation for most people. I dont know what the VRAM requirement for local generation is but it will be more than Illustrious.
VRAM requirements for training locally seem to be 24GB but I was training ILL locally with like 13GB.
RE: But it's cool that you're so quick with getting the ZIT lora ball rolling, I remember u were like the first with ILL too.
YES! I need to stay ahead of the curve. Plus its fun experimenting with new things.
RE: Re missing replies, no worries. I don't remember either ^_^
Thanks! I worry about it because there are some people that stopped talking to me because I dropped the ball 😵
RE: I hear the ZIT team officially requested the database material from NoobAI's team (but it seems they were refused?
WOW thats crazy! I thought all the data they uses was just this https://huggingface.co/CyberHarem/datasets
Dang... I was just thinking about building an app to make my own version of this multi-million image dataset but I figured there was no point. Maybe there would be a purpose in making that app. I guess I should buy that 1,000 TB hard drive 🤷♂️ Lol I can probably make the software to get the dataset but I dont have any way to store or use it.
RE: I see Riju's been joined at ZIT by a bunch more of your waifu
Yep! I have at least 40 ZITs posted now. I'm trying to keep an open mind but I still prefer Illustrious because I can make exactly what I want. (other than realistic things)
I think if I could train locally I'd be able to make better ZIT loras because I wouldn't be limited by CivitAI's number of training steps restrictions.