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
Improved general quality of the Model
Big Update with RD Photo 2
FAQ
Comments (57)
Exciting, can't wait to try it.
If I understand correctly, you expect that you'll be able to use all the training datasets you created for SDXL for Stable Cascade?
That should be highly convenient and good for faster adoption than SDXL has had.
The plan for now is to train the basic set of JuggernautXL. That's about 5k images, and additional content will come in further updates. As with JuggernautXL version 1, the focus initially is on creating a solid "foundation" for a Cascade Finetune Model
Regarding my own matter, I didn't want to post this directly in the description but rather make a small comment about it.
Of course, I haven't missed the "hype" around Juggernaut in recent months, which was ultimately evident from the numbers and image generations on CivitAI. (A big thank you, by the way, for that.)
With this comes naturally a greater pressure for me to always improve in order to provide you with continuous enhancements. Also, to meet your expectations of me. I want to mention that I'm just a regular "Average Joe" like most of us here, not a multi-million-dollar company. So please forgive me if the progress may not be as significant as theirs.
However, you can trust me when I say that I strive to get better every day and share these results with you through these updates :)
You've provided beyond my expectations already.
Started testing version 9 today and so far it is great. Thanks for constantly providing these great models and consistently improving things.
This model version brings all other models on Civitai to shame. I salute you Sir.
"Hell yeah, new Juggernaut, I'll just try a few things before I go to slee - oh fuck it's 6:30 am"
thats me since i discovered stable diffusion and try every model :(
@stevenlemke88948 pshhh who needs sleep or free hard drive space
If you are going to "reboot" for V10, try to bring back some of the SDXL base aesthetics.
I started using Juggernaut XL from version 5, but I recently tried V1 and V2, and the SDXL base as well, and I liked the aesthetics a lot, but there were too many quality issues.
The currently trained initial Base Set of JuggernautXL was captioned with Blip2 back then. While it was somewhat okay, it also contained a lot of nonsense and sometimes wasn't precise enough.
Since V5-6, I've been doing the captions by hand or using GPT4 for them, and since then, prompting has not only become a bit easier but the quality has also noticeably improved.
The goal with the reboot is simply to raise the basic structure to the same level as the last update sets. I'm hoping for another leap in quality... both in aesthetics and in prompting
Wow, keep up the good work, really looking forward to your stable cascade version. Glad the community is always ready to embrace the next thing.
Even though I’m just starting, I have to say this is the best realistic generation model, thanks a lot for your work, keep it up!
Using GPT-4 Vision to generate image descriptions for the training sounds like a great idea, but won't you need some kind of fallback (like LLAVA) for images which are NSFW? OpenAI is... rather unbending on this point. :(
In the future, I want to have a clearer distinction between SFW/NSFW content. This time, the base set will be trained without any NSFW material. The NSFW part will be trained separately in an additional model and then added to the base set. This way, I can ultimately present both: an SFW version (more suitable for businesses, schools, etc.) and an NSFW version for Civit or you all. Indeed, the NSFW model will be captioned using a different tool
@KandooAI Sounds like a very logical approach 👍🙏
@KandooAI Does this mean v9 is trained mostly on SFW images?
@uncreativ No, V9 was trained on SFW and NSFW, no changes here.
especially llava 1.6 33b is really good for pictures description ...gpt-4v level
Indeed, the generation quality of the character portraits has been very good. However, so far it has not been possible to obtain the correct male genitalia.
does this model produce images with an invisible watermark?
No
Good job on v9. For your cascade training would highly recommend adding a little nsfw stuff and even yoga positions and gymnastics for understanding of human body more to make the sfw stuff better. I did some testing on the base cascade model it's far more censored than SDXL was and it's making even non-nsfw things a lot less interesting.
If you do heavy negative prompting you can basically get the same breasts front on camera without an aerola and it's not very good, sometimes a side angle view of it and has a very cgi look. When you add any sort of posing it disappears. I feel like this could make it harder to get anything even remotely "sexy", I have concerns the base model is too censored and that even after training it might not work well as well as SDXL, but hopefully I'm wrong.
Just download every single NSFW heavy checkpoint on CivitAI and smash them all together at a 0.3-0.5 multiplier in A1111 and you're golden. Look at my profile pics, those were all generated by smashing every NSFW checkpoint I could find.
@user1234123 Yogo and Gymnastic Positions are a thing i def train on a SFW Model :)
If you guys want a very high quality NSFW Juggernaut, just download every single sdXL checkpoint on CivitAI that has "NSFW" in its description. Then smash then all together at a 0.3-0.4 multiplier in A1111. When you're done, smash Juggernaut with that beefcake of a checkpoint you made, starting at 0.3 multiplier then keep adding for each new version until you get the result you like. Juggernaut massively improves the low quality of the NSFW heavy checkpoints here. Keep fine tuning it and eventually you'll find a winner. It took me 25 merges to find my winner.
Dont ask me about uploading my ckhpoint though, my internet is too slow, its so slow that CivitAI rejects the upload if I even try.
EDIT: I assume the large amount of thumbs downs are people with potato pcs that cannot do SDXL merges, just simply too lazy to do it themselves, or your internet is super slow and downloading 100+gigs of checkpoints is out of the question. If thats the case and you're looking for a NSFW realistic model i recommend SDXXXL or RsmPornXL - Embrace The Suck (Beta). Both incredibly good and will only get better.
You could send to me directly using something like Megadrive or googledrive and I can upload it. no problem. Unless you do not wish to share then I will understand.
@zGenMedia Its good, but I'm a perfectionist. When im satisfied that it can generate a perfect vagina without the help of loras, then I will attempt to upload it for ya. Currently its too dependent on loras. Before the fine tuning the loras didnt even help
@fox23vang226 Look into SuperMerger so you can pick your favorite LoRAs and merge them into the checkpoint. My suggested workflow would be like: [NSFW Checkpoint A] + [NSFW Checkpoint B] + ... [NSFW Checkpoint X], whichever you were merging in already, then when all smoshed together, take your NSFW Checkpoint Merge, merge in the LoRAs you like at LOW STRENGTH (this is important), then merge the resulting checkpoint with Juggernaut.
@fox23vang226 count me in ;)
dude put that hog of a model on a thumb drive, go somewhere with better internet, and upload- you'll be a hero overnight
@zGenMedia https://drive.google.com/file/d/1yyVjy76UdxEr6yXnvRd7fSQC6FM36289/view?usp=sharing
Had I posted it myself this would be its description- but since you post it feel free to do anything you want with it. I will be uploading the example pics(https://civitai.com/posts/1559701) in my profile labeled Poppies NSFW mix. If you feel like its not up to your standard and dont want to upload it thats fine. I'll keep a link of it up on my profile anyways for anyone interested.
---------------------------------------------------------------------------------
Title-Poppies NSFW mix
Description-
Im a degenerate, I use SD mainly for corn. Mostly softcore corn is my style. In my quest to merge a model with a good default vagina without the need for loras, Ive come to settle upon this checkpoint. Although using NSFW loras as an addition does help further the realism. Ive only strictly tested this checkpoint on softcore corn, nothing more extreme.
Its not a perfect merge, you still have to cherry pick and get deformations in 1 out of 3 generations if you dont use controlnet. But I like this model a lot and used a variation(older version of it) in most of my recent profile gallery pictures. Ive generated a thousand pics with this model and even though its finicky I like using it for its consistency in the female genitalia region(further improved if used with a vagina lora).
https://civitai.com/user/fox23vang226
I highly recommend using hires fix. If you want to replicate my generations check out the prompts in my gallery, they've used almost exclusively a version of this checkpoint.
If you get X rated hardcore things showing up, just put penis in the negative prompt.
Credits to the creators of SDXXXL + RSMPORN +ZavyXL + Juggernaut (terribly sorry I cant remember the other checkpoints I used). Im sure there is a way to pull the metadata from the checkpoint, but I dont know how. If anyone does, feel free to share it in the comments and I will give credit were its due.
The "Fixed" SDXL vae is baked in
I am unable to upload large files to civitAI for reason unknown, the upload keeps crashing so I had to upload to Googledrive.
Verry impressed with versatility of model, great stuff and stunning visuals can be achieved by simple prompt only, no need for gazillion of fancy lora's, embeddings.
Great model!
Will you be posting a model card on huggingface.co???
Safetensors is up already. Link in the bio. Diffusers is on the way.
@RunDiffusion Fantastic. you guys are the best.
I just ran an image and I will tell you I was very impressed. Now to see what this will do in a full image. I strive for realism and this model just might deliver.
I would love to get to the point where the negative prompt needed for good results wasn't the length of a Hemingway story.
You don´t need a long negative Prompt (sometimes you dont need any neg prompt at all). I simply use mine from the good old 1.5 days cause i am lazy :D
@KandooAI You're right that sometimes you don't need a neg, but I do test models with author negs and none and definitely see a difference. I know that baking the negs in can be dangerous for creative dynamic prompts, but it's something to consider.
For some reason I can't generate any images with version 9. I try to generate four images at a time but your new version won't even generate one image.
V7 had best skin texture imo, looks to me like 8 and 9 bias toward plastic/airbrush feeling
judging by sample/submissions I should state that I still need to try my own prompts
@mumbles You could add textured skin, film grain and things like that to your prompt
I thought the same thing, I just tried 9 and everything looks very plastic compared to 7 and 8 (8 still looked fine for me).
Wondering if it's the included RunDiffussion Photo2 doing that? skin was so much better before
i am trying to amke an app that will run on stable diffusion , am i allowed to use this model in my app or do i need the devloper permission
Sure you can :)
As long as you respect the license : Crediting the Creator
If you don´t wanna put the Creator (in that case me lol) Name in your App (for whatever reason) than contact : [email protected] and we will find a proper solution for this :)
How about a round of Juggernaut Lightning? What do you guys think about that?
I'm already using it with Lightning LoRA.
I appreciate the gesture but I wish all checkpoint creators would stop baking LCM and now lightning into their models; its unnecessary bloatware. People need to learn to just use the LCM and Lightning loras, its far less bloat and far more flexible.
@fox23vang226 It's not like anyone is being deprived of anything. Personally, I think the regular V9 checkpoint is significantly better, and I also believe that we should be working directly with LCM&Lightning LoRA´s. However, for 2 months now, some people have been asking for a faster version of Juggernaut, and 'Lightning' seemed like a good option here :)
The regular versions of Juggernaut will of course still be coming (that's more my area anyway :D)
@KandooAI @fox23vang226 I agree that using the lightening LoRA is the better option, but many people are just confused by all these options. Should they use Turbo? LCM? Lightning? What do all these LoRAs do anyway?
Having seen how many people can't get normal style or concept LoRAs to work even if their lives depend on it 😂, I think having a version with Lightning integrated into it is a good idea, specially since KandoAI has already done the test and determined that Lightning produces the best results for this model.
@NobodyButMeow > I agree that using the lightening LoRA is the better option, but many people are just confused by all these options.
Of course, people who have trouble reading and understanding the line "download a 400Mb LoRA file and use with any regular SDXL model" are better off waiting and downloading multiple 6Gb files with lightning-version of models, it's much easier.
^^^ sarcasm
@tigerart You would be surprised how clueless people can be. But hey, we are all stupid/clueless in some ways 🤣
Loving it so far! Able to push resolution pretty high too. 1600 X 768 no problem. On Sampler you said "DPM++SDE"... does that also include any samplers that only start with "DPM ++ SDE ____ " ( Karras, etc.) (Automatic 1111.
Details
Files
juggernautXL_v9Rundiffusionphoto2.safetensors
Mirrors
juggernautXL_v9Rundiffusionphoto2.safetensors
juggernautXL_v9Rundiffusionphoto2.safetensors
juggernautXL_v9Rundiffusionphoto2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
juggernautXL_v9Rundiffusionphoto2.safetensors
juggernautXL_v9Rundiffusionphoto2.safetensors
Juggernaut-XL_v9_v2.safetensors
juggernautXL_v9Rundiffusionphoto2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
MyBack_SDXL_Juggernaut_XL_VAE_V9.0.safetensors
JuggernautXL_v9_RunDiffusionPhoto_v2.safetensors
juggernautXL_v9Rundiffusionphoto2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
model.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
mbrhan92.safetensors
mbrhan92.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
juggernautXL_v9Rundiffusionphoto2.safetensors
juggernautXL_v9Rundiffusionphoto2.safetensors
jx9.safetensors
juggernautXL_v9Rundiffusionphoto2.safetensors
juggernautXL_v9Rundiffusionphoto2.safetensors
Juggernaut-xl_v9.safetensors
juggernautXL_v9Rundiffusionphoto2.safetensors
sdxl1.0_juggernautXL_v9Rundiffusionphoto2.safetensors
juggernautXL_v9Rundiffusionphoto2.safetensors
juggernautXL_v9Rundiffusionphoto2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
juggernautXL_v9Rundiffusionphoto2.safetensors
sdxl1.0_juggernautXL_v9Rundiffusionphoto2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.