∞ 218 SDXL aka 'counterproductive illustration' omE

I only recently started dabbling in SDXL models, but I already put a lot of testing into it. Mostly comparisons, testing loras, trying to fix some shortcomings in models I liked.

I decided on a model called ReproductionXL because it looked the most similar to what I was used to, but better. After that I merged CounterfeitXL because Counterfeit aesthetics are still one of if not the best. Then I merged that result with Anime Illust Diffusion XL as I found the idea of being able to use different artstyles with tags fun -- and it is a lot of fun.

This took around 2 days of manual scoring and guidance through a Bayesian process, usually 500 samples per merge, which I did about 7 times total to arrive here. Also merged a LoRA trained on 18000 images at .4 weight into the final result, aimed at fixing sdxl models' lack of knowledge about the 4 letter word.

And so there you have it, that's the current point in time.

Onwards I'm looking to see how I can use Lynn's Animagine XL 2 and upgrade the main ingredients with their latest updates. I'm looking to add multiple characters (animagine should help with that) and more styles that aren't currently present within the model.

12 27 2023

• Converted to diffusers
• Uploaded neg embeds to misc
• Test merges in test folder

11 29 2023

• Yoinked layout
• Added images
• Uploaded 218xl

∞ Usage

So this is a bit different compared to usual models. You can still prompt as normal, but there is a fun quirk thanks to Anime Illust Diffusion.

There is a wildcard text file in the repo that contains all the tags that call for their respective artstyles, try and play around with that.

You can call it with __artistsAI__

There is also a wildcard with artists that I personally liked in this model, __artists218__, but this is a work in progress.

Artist xyzs thanks to b74 from the Touhou discord

One

Two

Embeddings Downloads

• aidxlv05_neg, Direct download
• unaestheticXL_Sky3.1, Direct download

VAE

• sdxl_vae, Direct download

∞ Credits

• Checkpoints

  • ReproductionXL v8.7
  • CounterfeitXL v1.0
  • Anime Illust Diffusion v0.52

• Lynn for layout and future juice

  • Animagine XL 2

• Ashtaka Aha Oof for taking my spam on discord EZ Clap

  • Ahnn

• Tools

  • Bayesian merger by s1dlx
  • LoRA_Easy_Training_Scripts by Derrian

∞ Legal deez

You are free to use this model locally but
1. You aren't allowed to redistribute on another platform. (like CivitAI or Tensor.Art)
2. I am not responsible for how this model is used to generate images.