Welcome to Rets' Hyper Cheek Clappin Merge_v1 (TM) :)

Hello Civitai community! You don't know me, and I probably don't know you (yet :D)! This is a merge I've been working on for a month or so now. I'll be honest, I've only just gotten into merging and so there may be some refinement and polish that needs to (or will) happen with this model.

The goal I set out with was simple: I wanted a model that had a broad conceptual understanding, solid anatomical representation, understood both natural language (to some degree) and danbooru tags, and rendered a variety of anime styles with reasonably high detail. I think I've achieved most of these here, though I am ALWAYS open to suggestions or tips to improve!

So for what I did here:
I started with an Illustrious 2.0 Stable Base, and merged several models that I have used and loved into it, using block weighted merging and a fair bit of testing. Some of these (the ones I actually wrote down and remember) I will list below. I went for a mix that would give good anatomy, the anime style I personally find appealing, and a solid conceptual understanding to minimize the need for LoRAs to achieve certain results and aesthetics.

I then re-merged the end result with an earlier stage of the process to reinforce the stylistic fidelity that I was after.

Then came the LoRA merging stage, where I reinforced some of the concepts I have personally used that I needed LoRAs to achieve in the past. I kept the weights low for all style and concept LoRAs to keep the effects subtle and not change the art style too much. I then merged some quality LoRAs to enhance details (background, eyes, hand and feet anatomy, shading and color). After this I re-merged once more to correct style drift with a model to keep the art style. This may not be the correct method for merging, but its what I decided to test here.

The end result is a model that should understand both natural language, danbooru tags, and respond (subtly) to artist tags. It should baseline have a strong 2D anime style with fine linework and good detail. It works with almost every sampler/scheduler combination I've tried or that seem popular, and the results are available of this testing in the example images. It has pretty solid anatomy and conceptual understanding, so it should be versatile with a myriad of flavors of art. It does not do realism well, as that was not my intention with this model. It also does become a bit unstable at ultra wide aspect ratios with regards to anatomy. This is something I am working to correct, but it is still an issue at this time.

It should work exceptionally well with wildcards, generate character model sheets with multiple poses and angles, and cleanly represent emotion or mood in images. Its lighting and shadows are also quite good, so images should feel quite cinematic if prompted for correctly. This model is equally capable of SFW and NSFW images and concepts. Use related tags responsibly in your prompts.

I hope you enjoy this model as much as I enjoyed making it, and look forward to more and updates as I learn and grow from feedback and testing. Feel free to share your works and comments, I will do my best to read each and every one of them.

Recommended Settings/Parameters are as follows, but feel free to experiment! As stated, this model responds reasonably well to a variety of samplers/schedulers and settings:

Sampler: Euler/DPM++2M SDE/Euler A

Scheduler: Beta/Karras/Normal (Automatic for your A1111/Forge users)

CFG Scale: 5.2 - 7.5

Steps: 24 - 40 (37 seems to be a sweet spot, as well as 28, at least in my testing, and depending on sampler)

Recommended Resolution: Varies depending on your prompt or intended image, but I've tested all of the usual suspects, 832x1216, 896x1152, 1024x1024, 1152x896, and 1216x832.

I HIGHLY recommend hires fix or iterative upscaling, but understand that some machines won't be able to do this. All of my testing and my uploaded images used iterative upscaling.

Now for the models I KNOW are in here:

::Checkpoints::

• Amanatsu - https://civarchive.com/models/1325426?modelVersionId=1496755

• Conceptor - https://civarchive.com/models/1637646?modelVersionId=1853705

• ILLv - I Love Lucid Visuals - https://civarchive.com/models/1291368?modelVersionId=1791659

• Miruku - https://civarchive.com/models/1442151/miruku

• Plant Milk Model Suit - Walnut - https://civarchive.com/models/1162518/plant-milk-model-suite

• RAMTHRUST'S-NSFW-PINK-ALCHEMY-MIX_v012 - https://civarchive.com/models/1465491?modelVersionId=1682833

• Soft Sketch Illustrious XL - https://civarchive.com/models/1556823?modelVersionId=1761681

• Unholy Desire Mix - Sinister Aesthetic (Illustrious) - https://civarchive.com/models/1307857/unholy-desire-mix-sinister-aesthetic-illustrious

  :: LoRAs ::

• INCS - https://civarchive.com/models/1064940/incs

• HORRORCORE | Experimental LoRA - https://civarchive.com/models/1415639?modelVersionId=1600054

• Space backround LoRA [XL+Pony+Illustrious] - https://civarchive.com/models/552983/space-backround-lora-xlponyillustrious

• Monster girls Lora [Illustrious] - https://civarchive.com/models/1558989/monster-girls-lora-illustrious

• And some quality and detail related stuff but i cant remember exactly which ones (the most common Illustrious trained LoRAs for these things)
  
  All credit and respect and admiration for each of these models goes to their creators, without their hard work this model would not have been possible.

  A special thanks goes to Cgibson and Ramthrust for their assistance in learning the ins and outs of ComfyUI which i used to create this merge.
  
  All base model legal use and licensing rules apply to this model, and this model is not for upload or use on paid site generators other than CivitAI.
  
  And as always, you are personally responsible for your outputs and generations. Please create responsibly. But above all else, have fun!