Damn, I would never have imagined a combination of Lustify with Juggernaut

New version is so good

I feel like someone really needs to put the time in to do a good 3.5 large finetune. I had thought Juggernaut was being worked on for sd35?

SDXL really was the GOAT for fine tuning.

Good new model and very polyvalent.

Nice work!

Holy f*ck, I didn't expect that YOU would use my work as a part of the base. This is an honor for me

Amazing SDXL model again. Thank you for your effort and evolution

Amazing. This will be my new favorite, replacing Juggernaut XI.

I've been waiting on this one. Thank you for your hard work.

Anyone having luck with DMD2 Lora? im getting frequent deformities. Have not tried it without DMD2 yet though.

Hmm, isn't letting me use it. Says dormant and unable to be used to generate and to contact Civitai staff. Weird!

Pony7 for sure is the next step. You must collaborate with them. Big name, first finetune. Good for everyone.

Big censored models are boring, v7 is the only one, im exited about.

Hi, thanks for the new model! I use Juggernaut all the time for image generation. It’s sad to hear that nothing fires you up anymore, but I do agree that open-source models are in a bit of a slump right now.

I still break out the 1.5-series models for certain jobs, because nothing else handles extreme resolutions properly.

As for GPT: at first I was an impressed, but by the very next day I’d already hit its limits. You’ll never get truly epic backgrounds with GPT, and you can’t train a closed-source model on the exact style or concept you need. Even when you feed it references, you just end up with that same damned “GPT style.” And forget about 16:9

Midjourney, meanwhile, can ban you on a whim if it doesn’t like something. I’m not even talking about porn—horror imagery is off-limits too, as are dark, brutal covers for brutal death-metal or goregrind bands.

In the end, the only real strengths of closed-source models are pretty portraits, cute cats, and memes. I’ve got nothing against people pumping out memes, but for serious work closed-source models’re very frequently useless.

Don't worry about Flux, Sora and similar models. All models are good and are used for different purposes.

I’ve already tested Ragnarok, and I can tell you it’s fucking awesome.

Thanks again for your work, I hope you find inspiration for your self-realization in the future.

Thank you, your work is a source of inspiration to me, can't wait for your next project.

I hope someone can help me out here. I am running Juggernaught XL on a paperspace cloud machine P5000. I have run the following prompt 15 times with no good or nearly accurate results or even close. 1 fully body bad pose, 10 just legs in heels most have deformed legs and toes. no facial details like moles or biting of the lip. 3 portraits disfigered with no torso. If someone can please help me understand why I got no good results after so many attempts? I was under the impression Juggernaut is the best model that is around for both realist models and NSFW?

Prompt: A hyper-realistic full-body portrait of a 20-year-old athletic woman, (highly detailed symmetrical face with flawless skin, centered in the frame, head fully visible:1.3), long straight strawberry blonde hair cascading over shoulders, large almond-shaped blue eyes with long eyelashes, (biting her lower lip seductively:1.3), (distinct mole above the right upper lip:1.2), (toned athletic body with smooth curves, anatomically correct torso:1.3), large breasts in a delicate red lace bra, wearing no underwear (extremely tiny, barely covering the pubic area, exposing a subtle strip of visible pubic hair:1.3), (sitting on the edge of a white linen bed, legs slightly spread in a clear frontal view, head and upper body fully included:1.3), hands resting on the bed sheets, wearing glossy black high stiletto heels, natural soft light from a nearby window, ultra-detailed, photorealistic textures

Negative: lowres, blurry, jpeg artifacts, watermark, text, signature, deformed, poorly drawn, mutated hands, mutated limbs, extra fingers, extra limbs, extra arms, extra legs, bad anatomy, bad proportions, disproportionate body, unbalanced limbs, disfigured torso, distorted torso, unnatural torso, missing head, head cropped, no face, faceless, distorted face, unnatural facial features, censored, doll-like, cartoon, anime, CGI, plastic skin, overexposed, underexposed, double exposure, extra torso, double image, cropped image, out of frame, unnatural lighting, unrealistic facial features, missing mole, incorrect hair color, clothing errors, high-cut panties, fully covered pubic area, missing pubic hair, neutral facial expression, closed mouth, missing lip-biting, standing pose, lying down, incorrect pose, no bed, missing bed sheets, anatomical errors, distorted proportions, lack of realism

DpM++2M
Sampling step 40
hires step 6
Denoise 3.5
upscale 1.6
cfg scale 9
seed -1
512x1024

Having hard time trying to get something out of realism, like sketch or drawing style,even if I put realistic, realism, photoreal in negatives

Thank you and your team for all the efforts you put into this. Juggernaut has become one of the leading Models for SDXL - at least it was for me.

I also never got warm with Flux as it is horrible with work with. So yeah, thank you again for Ragnarok.

I won't say "farewell", instead I wish you all best of luck and hope the Adventure of Juggernaut will continue in some epic way.

Again, thank you and keep it up! You guys rock!

Very nice so far!

I can't seem to generate small nipples/areolas though. Even with weighting or LoRA.

Anyone else have this problem?

you are an angel. !!!!

Has anyone made an inpainting model?

Hi, I've always used all your models, I love them, and I can tell you that they are the most creative models I know. FLUX stands out as a model only because of the good hand correction, but it's a very closed model in terms of creativity...

Keep going and don't forget that these models have made people recognize your good work.

Congratulations, friend... :)
Thank you very much...

You've made fantastic SDXL model. Love creating with it!

Used this model on my 3060ti, crashed my PC inexplicably to the point where I'm scared to ever touch it again: Fair enough.

Remote desktopped on my friend's 4070 SUPER, and whatdya know? CUDA error and a crash.

Respectfully, what kind of GPU do you need to run this? 🤣

Kinda cool what you can do with it. Nice.

How come I'm able to use Illustrious and Pony LoRA(s) on Ragnarok? or is it just a placebo?

one of the best right out of the box. thanks for charing :)

Anyone else having issues inpainting w/ this model? Applying the Fooocus patch doesn't really work at all for me.

Also tried merging it into a pseudo-inpainting model via the "Add difference" merge (if you know, you know), but that fails to give good results as well.

Both of these methods have always worked with JuggernautXL models in the past. Anyone had luck with inpainting?

That’s incredible model that requires very little prompting to get the right results. But I find it very unresponsive to the LoRas. Same setup with Juggernaut below v.10 follow the loras precisely, but higher than 10 and especially the current one feels like totally ignoring the loras. Is that expected behavior?

I am new to sdxl I normally use flux model but trying other models. What is the difference between Jugg_XI_by_rundiffusion vs ragnarok_by_rundiffusion? the older one has such good reviews but notice ragnarok being newer is it better?

chocolat

Woo

...and the plot thicken...

I installed open-webui last night, generated images using v13 using few lengthy prompts, and somehow, a prompt with 203 tokens still adheres?

Seems like model comprehends complete sentences properly than comma separated tags.

NOTE that this doesn't feel creative. It is hard to manipulate images in this way.

The Ragnarok Model is awesome!!! Thank you very much!

The sample images don't seem to show up anymore for any version... I've checked other checkpoints and the issue is limited to the juggernaut page.

Have they been removed? Is it a bug?

For the lightning version, what sampler should I use, and if it's not one I already have where will I find it?

Is juggernautXL_ragnarokBy suitable as a base model for LoRA training?

Do you have any suggested inpainting workflows for Ragnarok?

The checkpoint Ragnarock has no idea what furry/anthro is.

But furry LoRAs are working as intended (at least, mines do. Tried with the six of them : All characters are drawn correctly).

Ragnarok version is one of my top3 favorite checkpoints, with CyberrealisticXL. It's very versatile and i like how i can combine my weird prompting style, sometimes long sentences with shorter random words and terms like "memories of past" and it does cool stuff with my flow of mind type of prompts that do sometimes. Thanks for your efforts and good work!

Much love to you all, wish you have an amazing warm summer day. You all live in my heart , in my prompts! <3

I like this model, beautiful images. There is sometimes a problem with eyes, I find eyes are created without pupils. This is terrible if you are trying to get images together for a model.

It would be nice to find a checkpoint that can deal with underarm hair and not go into cheesecake poses of arms thrown behind the head.

Damn all this and just fp16

It's off the generator :( nooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo

A huge thanks to the creators of this amazing checkpoint! Resolution 4096 with detailed textures. This is just fantastic! With the help of RTX 3060 in 3 minutes of generation.

my problem is that the eyes of my pciutures are deformed. how can i fix it?

Animal fur has never been better in SDXL. Apart from generating, I use it for i2i in SwarmUI with great results to improve animal pictures with less than optimal fur rendering. For that purpose, Euler even does a tad better as sampler, being less aggressive as to animal fur texture than other samplers.

Can someone please tell me why this won't make a P*nis that doesn't look like a Johnsonville Bratt!!

Wow!

Really nice general usage models. Can make all sorts of stuff. I'm partial to V7, seems to generate stuff in the most artistic and most aesthetically pleasing way, at least to me.

An exceptionally realistic checkpoint that sticks to prompts and is able to generate characters as requested (regardless of race or features) with two downsides:

1. Sub-par representation of hands. I get far, far more extra fingers or missing/botched nails than most other modern checkpoints.

2. The facial features are very repetitive, resulting in a low variety of generated "humans". Using elaborate wildcards that create a substantial differentiation of prompts, creating 1000 images, after the first 100 I see mostly repeated faces.

I like this model (especially Ragnrok) and your work in general, but I don't think hiding negative comments would do you any good. on the contrary.

Nice rendering but strangely trained model, lots of logos, text and brands... Not sure it is safe to use on production, because many final pictures have similarities with (c) ones... Other models create original art. This one not so much. Maybe this needs a fix.

Is there a lightning version of Ragnarok?

With so many versions, which one should I use? Is newer always better?

Wow! This Is Beast Thanks.

Seems to be a good balance between realism and flexibility

I`m a new user with comfyui, can someone help me?I try generate 1 output, no matter what I choose the resolution (256*256 512*512 768*768 1024*1024), my generate speed is be fixed in about 7~8 it/s, the gpu usage less than 30], it trouble me a lot.
I am using 4090, Python version: 3.12.7 Total VRAM 24563 MB, total RAM 49070 MB

pytorch version: 2.7.1+cu128

xformers version: 0.0.31.post1

Set vram state to: NORMAL_VRAM

Device: cuda:0 NVIDIA GeForce RTX 4090 : native

Using sage attention
I don`t know if the pytorch version reason,cuz I use the original comfyui : 2.6.0+cu126 can generate the normal speed (13 it/s in 512 *512)

Thanks

Good

My favorite model for SDXL, excellent quality, consistent results, capable of inpainting and controlnets. Overall excellent.

0.1 and 0.15 .. you mean like naive full merge, not blocks anything?

The goat, clean textures, works best around CFG 5. Good detail in portraits.

I was asking google, what checkpoint is less artsy & I was directed here. But of these listed, which one would make a less artsy wolf?

informarles de que este modelo está presentando errores y generando resultados incorrectos actualmente.

Why can't I see this base Juggernaut Ragnarok base model to choose and use any more? Am I banned from using this as base model!?

hiii thx!! one question tho, how do you install HiRes: 4xNMKD-Siax_200k with 15 Steps and 0.3 Denoise + 1.5 Upscale?

I'm pretty sure I've found something here. I'm impressed with the images I've got using this checkpoint. Thank you, great work!

the model has his temper... but when you tame it goes a long way, I'm using it to restore some old photos, and ragnarock it's making magik

This model is amazing—I set the sampling to 150, kept the default width and height, and left everything else unchanged. Even AI can’t detect the level of realism it produces. I’m running it on an RTX 4060 with 8GB of VRAM. you are great author

Just the Best

May I ask which VAE this model uses? I can't get the correct images when I use 'sdxl.vae'.

I think I'm missing something. I'm trying to use it on simple prompts for generating pin-up female models in old fashioned style, but it constantly gives me lots of deformed eyes, hands, extra fingers.

A finetune based on z-image or klein is calling your name!

I'm terribly confused because there are a lot of models here to download with different and confusing names. WHICH ONE IS THE LATEST?
EDIT: had to ask LLM. Last one is "ragnarok" and yes, its not a "special version" its the successor of jugg XI

What schedule type should i use?

amazing job

Its a good model. I posted an image I always do when testing a model because it will show contrast, lighting, reflections, hands and eyes. Also good is it's a trained model and not merged model. For me, its a keeper. Thank you for your work and sharing it

I just ran a bunch of images of the woman in the white sweeter and all of them produced perfect hands. This might be a good model to train a Lora for hands

Its Apr 2026, and Jugs Ragna is still my Go-to model! I'm looking forward for an update, or your flavor of flux.2 open weights.

Thankyou for all your effort. You inspire me and alot of others to make beautiful models like this one and make this community an even brighter place =)

Nice

hahaa

best Model File