Updated the Prompting Guide
For business inquires, commercial licensing, custom models, and consultation contact me under [email protected]
Join Juggernaut now on X/Twitter
Juggernaut Ragnarok on RunDiffusion
Juggernaut XI & XII on RunDiffusion
Prompting Guide for Juggernaut Ragnarok by Adam
Prompting Guide by Adam for XI & XII
A big thanks goes to RunDiffusion and Adam, who diligently helped me make it work :) (Leave some love for them ;) )
Hey everyone,
It’s been 8 months since the last version was released here on CivitAI.
Of course, I haven’t been idle during that time . I completed several projects to ensure I’d have the financial means to keep exploring new architectures and possibly do full finetunes on them in the future.
Juggernaut Flux (and its many sub-variants) was a ton of work, but ultimately, I’ve wrapped that chapter up. The training process gave me way too many headaches. To keep my sanity, I spent my spare time working on Juggernaut SDXL with the hope of maybe releasing one final version for you all.
And that day has finally come. :)
Juggernaut Ragnarok has improved in many areas: photorealism, digital painting, poses, hands, feet, and much more.
That said, it’s still an SDXL model, and I don’t recommend comparing it to models like Flux, Reve, or Sora. For example, it still has limitations when it comes to text rendering or faces at a distance.
I recommend using it as part of a pipeline for your projects. Example setup:
FluxDev / Pixelwave / Jug Flux Pro → Juggernaut Ragnarok
A quick personal note about Juggernaut:
Honestly, I don’t know what comes next.
After the release of Sora and similar tools, the open-source image generation space feels a bit dull in comparison.
Nothing has really excited me enough to dive back into training (yes, I’m talking about HiDream too).
So I’m seeing Juggernaut Ragnarok as a kind of farewell, especially since it’s unclear where things are headed with CivitAI in general.
(You can download all Juggernaut versions from HuggingFace, by the way.)
Last but not least:
Have fun with the model, share your creations, and good luck with your projects!
And in case you’re wondering: Yes, you can do anything you want with Juggernaut : merge it, train it, sell the image outputs, etc.
Just a simple shoutout is all I ask. :)
And now, here are the recommended settings:
Recommended Settings(VAE is baked in):
Res: 832*1216 (For Portrait, but any SDXL Res will work fine)
Sampler: DPM++ 2M SDE
Steps: 30-40
CFG: 3-6 (less is a bit more realistic)
Negative: Start with no negative, and add afterwards the Stuff you don´t wanna see in that image.
VAE is already Baked In
HiRes: 4xNMKD-Siax_200k with 15 Steps and 0.3 Denoise + 1.5 Upscale
And now, have fun trying it out. As always, I'm eagerly waiting for your pictures in the Gallery :)
If you liked the model, please leave a Like. In the end, that's what helps me the most as a creator on CivitAI. :)
Last but not least, I'd like to thank a few people without whom Juggernaut XL probably wouldn't have come to fruition:
Dreamlook.AI (Trained 3 Side Sets)
Description
220k Steps
FAQ
Comments (30)
sheeeesh
Finally, one of the heavyweights masters of SD has launched a XL model. Let's try it!
Don´t expect to much yet :D
It´s like with the Original Juggernaut. Version 1 on SD 1.5 was very clunky back in the days :D
Still lot to do with Juggernaut XL but we will get that perfect photorealistic look sooner or later :D
@KandooAI From what I've seen it's still quite good! You mentioned you're going to further enhance the nsfw aspects of it in later versions (I'm sure it's been a pain and a half with XL) Are you going to try to incorporate more complex actions in it as well with 2 people interacting (cunnilingus comes to mind as one of the more difficult things) or does XL make that pretty much impossible and we'll just have single characters for the most part until quite a awhile in regard sot nsfw?
@Light7799 Indeed, it's quite a tedious task to incorporate nudity into the model. The ultimate goal for Juggernaut is to perfectly depict both women and men naked. However, there are no further plans in that regard, as it would shift Juggernaut's focus in a direction I hadn't imagined ;) Regarding the question itself: I believe a "hardcore/More Complex" model would certainly be trainable. A good dataset is already half the battle. For a purely NSFW model though, it's work I wouldn't do for free :D It would likely be a classic closed-source project. However, I'm sure we'll see a larger model in this area soon on Civit ;)
Dunno, SDXL in general seems a little off to me. I wonder if the dual text encoder was a good idea on the part of Stability AI.
I am getting the "same" and in some cases better results from base XL. The same is with other trained custom XL models. I know that you say this is just the beginning. But working on it for now, what is your opinion on those improvements? How will they "occur"? XL is already trained on 1024x1024px. I don't get what will new training add to the dataset stability used? How does that custom training quality even work? Do you (or anyone else) somehow use higher quality images for training? 1.5 was a different animal, but with XL, i don't know.
@ledeni80 Stability's dataset contains billions of images. I would phrase it this way: SD Base can create many concepts well, but none of them are truly outstanding. At the end of the day, fine-tuning and LoRA's are responsible for that. Juggernaut's goal is ultimately to ensure good photorealism, but in doing so, it will "lose" or not be able to represent other styles as well (e.g., paintings). It's similar, or even identical, to SD 1.5. The base model was also a massive dataset, but it was ultimately the fine-tunes that unlocked SD's full potential.
I personally only use images with a resolution higher than 1024x1024 for training. However, the training process itself is a bit different (at least in my case). I'll be training more fine-tunes for Juggernaut and merging them into JuggernautXL. I adopted a similar approach with SD 1.5, but I think it's even better suited for SDXL. So, instead of training one large dataset, I'll break it down into several medium-sized concept fine-tunes and then combine them. We'll see if it all works in the coming weeks/months :D. However, I am quite optimistic that SDXL will deliver exceptional quality in a few months, not only with Juggernaut but also with many other fine-tunes (RV, Dreamshaper, and so on).
The only major issue I have with SDXL is its immense hardware hunger. This makes it challenging for the average user to utilize
@KandooAI That is one of the things that i was curious about. That "replacement" of data with the intended purpose of the custom model (in your case photorealistic). How your images influence their "billions" percent wise. How does base model "lose" data in your favor? I get the "fine tuning" concept in theory and in case of very theme specific loras, styles and custom models. But in case of a general photorealistic custom model, not really. Shouldn't there be among those billions of images a large number of HQ photorealistic photos? Could we say that that number would surpass your dataset that you intend to use even in all of the future versions combined? And by a large margin. Why won't photorealistic images from a base model "activate" without fine tuning? :) Searched for the explanation but i couldn't find it :)
wow, that's great news.
THANK YOU!!!
Finally the big guys are working for SDXL, can't wait for more refined models, it's gonna be awesome
Really excited about this. Thanks for your fabulous work!
hey,
great that you start now on SDXL models
may i suggest something, neg-promt leave blank until you do not want something
https://civitai.com/articles/1826/sdxl-prompts
i put also a image in your galery ;)
I agree on that. I am just lazy, so i just used my old 1.5 template :D It got the Job done :D
But yeah, XL works really good with a shorter neg-prompt
@KandooAI all okay, when do they decide on NSFW content? my work is fine in most cases and there are at least 6 other loras ;)
IT'S HERE. IT'S HAPPENING.
Can you train a V1.5 model on your sdxl images ? I want to boost some models with your work
Not in the near future. I simply dont have the time in the moment.
Right now i am working on a couple of SDXL Projects
Unfortunately all images I generate look good during preview, but turn blue or green corrupted as it finishes, seemingly no matter the settings.
The problem is the vae you’re using to decode the latent image generated. Use the suggested vae that comes with the model on this page (look where it says 2 files).
best trained XL model so far, bravo
An amazing model! Thank you for your hard work!
It is simple - the best!
i keep getting this error with your model, anybody know what it might be? "NansException: A tensor with all NaNs was produced in Unet"
P.S. i've already tried the --no-half command and the Automaticlly revert VAE to 32-bit floats setting
--no-half-vae works for me
I'm highly ask not to add another models in merge with next update. Juggernaut XL for now is the best one, please strenghten it by your own training ;)
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.