SURPRISE ;)
Merry Christmas, everyone.
As a little Christmas gift for you all, I've been working on a new version for the SD 1.5 version of Juggernaut in my spare time for the past few weeks.
I took a small portion of the JuggernautXL dataset and trained it on SD 1.5 (Base Juggernaut Final), while also updating the epiCRealism model in the merge. However, the RPG and Diva models that were added in the 'Aftermath' version have been removed.
Images were created with the following settings:
512*768
Sampler: DPM++ 2M Karras
Steps: 35
CFG: 7
HiRes Fix:
Sampler: 4xNMKD Siax 200k
Steps: 15
Denoise: 0.5
Upscale: 2
Lastly, the recipe for the last Juggernaut (Aftermath) version on SD 1.5:
SD 1.5 Base
Training Set 1 (0.65)
Absolute Reality (0.18)
epiCRealism V3 (0.2)
Humans (0.15)
Skin Enhancer (0.15)
Better Portrait Lighting (0.2)
Training Set 2 (0.55)
RPG 5 (0.38)
Divas (0.15)
And of course the NinjaFix made by @chillpixel
Thanks all those Creators for their hard work :)
Of Course you can run "Juggernaut" on Tensor.Art
Description
Inpaint for Aftermath Version
FAQ
Comments (41)
what is the difference between AFTERMATH and FINAL models?
he added some more data. See the about this version on top right of model page.
Are you going to do an update now that Absolute Reality and epiCrealism has gotten updates? :)
Well never say never, but at this point i´ll have no plans to work on SD 1.5 in the next weeks. All my resources and power goes into training Models/LoRA´s on SDXL right now
So, I discovered that "cheerful" seem to have overwhelming big toothy grin power, for some reason. That's pretty innacurate. Thought you should know.
Question. Does the prompting work better than in previous versions? This model is the most western and most realistic of the checkpoints.
Depends which older/previous Versions u mean. I restarted at Version 1.9/2.0 , everything prior to that was different and harder to prompt (in my opinion) . Since 1.9 its more Booru Tags driven (I trained my Set with WD)
hey Kandoo question about VAEs. I hadn't paid much attention as I was previously using stock models.
The final model vs the final model w/VAE. Adding your VAE seems to take away some of the contrast when using your VAE vs SD 1.5 VAE. Thoughts? While it's not a bad thing I mean I just prefer a higher contrast output.
Also, confusing to have a baked and non-baked model but not do the same for inpaint. Because it raises the question "does the inpaint need a VAE or is it baked?", which cannot be determined from information provided.
if you don't have a good gpu like me, you can try this on Google Colab (for free, no ban, no limited token). Also a useful guide here Youtube.
this is great! any chance for an SDXL version?
It's out!
the downloads from civitai have been extremely slow barely hitting 100kbps while files from huggingface gives around 2-3mbps. I've noticed it only pertains to civitai and downloading anything over 2gb is literally impossible
yeah takes me about 40 minutes to download something off of Civitai, it's been better these last few days, maybe try again, and @ me in the general discord what your results are @solaire
It gives white shade to photo when the vae is used so what's the point of using vae anyways?
Yeah I noticed this as well when playing around with different vaes. May be my lack of understanding but not adding a vae and using the one that comes with automatic1111 seems to result in the best output.
What is the difference between the 5GB file and the 2GB file?
Same question here... Why no reply from the creator?
What is the difference between the 5GB file and the 2GB file?
Same question here... Why no reply from the creator?
If you expand the box that says "3 Files" and look inside, you'll see the answer.
@SumRandomPerson can you just put the answer here? Cant see a box with "3 files"
best guess: the 5gb is unpruned. i'm getting nice images from the 2gb and it's fast, nice colors too
@ffffffffredo Totally miss the question. The 5GB Version is the Unpruned and the 2 GB the Pruned Version .
For just Image Creating the 2 GB Version should be fine. If you want to train on Juggernaut i highly recommend to use the Unpruned Version :)
What's the difference between the 5GB and 2GB versions?
The 2 GB is the pruned Version. Visually you shouldnt see a difference. The 5 GB is more likely get used for training. So if you just want to create images that you can go with the 2 GB Version :)
@KandooAI Aaaah, densmore, and more information packed? I don't want to pry.
I have finally gotten to evaluate the Aftermath version in depth and I believe it is a step backwards, and I'm sure I can pinpoint why. I have also evaluated RPG V5, and it is really not all that good in general, with especially terrible eyes on my setup.
Personally, I am rolling back to Final, which is probably as good as it gets. Juggernaut was always in my top 5 of all time until Aftermath, and while still very good, this version just doesn't elevate the series in my opinion.
I personally prefer the Final Version too. The Release was more to get something between Final and the Release Wait for Juggernaut XL . Also RPG5 was released this time and i wanted to give it a try.
Like i said, i prefer the Final also, but Aftermath was good enough and different to release it.
Hi, thank you for your awesome work. I have one question: How do you handle the image size if it is not divisible by 8? I think your output may not have dimensions that are multiples of the image size. I am using the Diffuser library for my Python pipeline.
Use a calculator and start doing 8x2 until you hit a numbers close to your desired width/height.
this is the best checkpoint for realistic clean humans, I tried many nothing beats juggernaut
The checkpoint is nice but the recommended VAE causes pretty horrible washout.
i really love this model as i can create realistic version of a lot of things! keep it up!
I always have respect for dedicated folks such as yourself.
Just a reminder; you may be able to get some kind of deal, donation, or discount, on compute time, if you contact relevant providers and mention you are doing non-profit work that is (this) popular.
Particularly if you mention you will give the company credit for the donation time, on the resulting model.
Or take a path that doesn't involve getting ass raped by a corporation for your time and stress, and freelance stuff instead. Corporations typically take about 50-90% of what you actually make for themselves. That's why so many people can barely afford to live these days.
i dunno why but face looks so washed out body is fine im also using VAE
when you say place config file next to checkpoint, how is that done, im a newbie using comfui so apologies if this seems a simple question
Don´t you worry about that Config File. In 99,9 % of the Cases its not needed. Just try it out without the config :)
@KandooAI but to clarify, it just means to place it in the same folder as the model before importing, no?
@Narz Thank you.
Details
Files
juggernaut_aftermath-inpainting.yaml
Mirrors
juggernaut_v2-inpainting.yaml
juggernaut_aftermath-inpainting.yaml
uberRealisticPornMergeURPM_urpmCE3Alpha.yaml
juggernaut_v21-inpainting.yaml
photomerge_v12-inpainting.yaml
photomerge_v10-inpainting.yaml
uberRealisticPornMergeURPM_urpmCE2Alpha.yaml
uberRealisticPornMergeURPM_urpmCE1Alpha.yaml
uberRealisticPornMerge_urpmv12-inpainting.yaml
animerge_v10-inpainting.yaml
juggernaut_final-inpainting.yaml
juggernaut_v19-inpainting.yaml
uberRealisticPornMerge_urpmv13Inpainting.yaml
juggernaut_v18-inpainting.yaml
juggernaut_v15-inpainting.yaml
juggernaut_v16-inpainting.yaml
juggernaut_v17-inpainting.yaml
juggernaut_v22-inpainting.yaml
juggernaut_v14-inpainting.yaml
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.