CREAT0R_MIX
By pAInCREAT0R
"Thumbs Up Reviews" are appreciated as it is the only way creators not in CIVITAI's creator program get their work noticed. - Thanks!
CREAT0R_MIX - The perfect mix of Illustrious, NOOBAI XL, PONY, and SDXL. Merged then trained to achieve Danbooru and E621 character recognition in a realistic style.
CREAT0R_MIX is a complex merge of 4 custom trained checkpoints (NOOBILLONYXL):
Why? So What?
Originally CREAT0R_MIX was being designed to be a realism refiner that could sit over multiple XL model types. The goal to bring more refined SDXL realism to the newer model types (PONY, ILLUSTRIOUS, and NOOBAI XL). A checkpoint that recognized the majority of the concepts when converting them to realism.
CREAT0R_MIX does work excellent as an all purpose refiner. However, testing showed that it outperformed many of the base models in realistic conversions of Danbooru tagged characters. It also puts some of the variety and artistic freedom of SDXL into the mix. If you find a concept it struggles with, try putting in a first pass with the base model type that excels in the concept and use CREAT0R_MIX for a second pass with a denoise of around .5-.75 or so (depending on model types).
How?
CREAT0R_MIX was made by dividing the SDXL blocks into 4 quadrants. Each block group was then filled with a custom merge of three of the four models. The earlier blocks were PONY, ILLUSTRIOUS, and NOOBAI XL. The later blocks replaced the PONY model for SDXL. The middle two quadrants were a careful transition merge of all four models. Block merging is nothing new or revolutionary - this combination though of merging different combined model weights into different quadrants allowed for many (MANY) combinations to be tested to reach the desired outcomes. Its not perfect yet - you will get some SDXL hands at times for example. However, it handles a large variety of prompts from other models well. It excels at rendering Danbooru characters in realism aesthetic.
V2 - Created by finetuning the original model V1 then doing a slight regression merge of NOOBAI to preserve concepts.
Recommended Settings:
Works good with multiple sampler/optimizer combos. Most of my images are using the following settings:
35-50 steps
Euler-Ancestral (Euler-A), DEIS, DPM++2M
Beta, SGM Uniform, or Normal
Upscale - HiRes fix at moderate strength (.25) with a flexible upscaler works best. I do not use Adetailer, but I know some people like it. I always struggle with it in COMFYUI, but I did like it when I used A1111.
Early testing seems to show Illustrious LoRA's work best with the model. This has not been widely tested yet.
Note: I did not intend to have this many models in circulation. Once I had a plan to make this model, I needed to train THRILLustrious and NOOB_AIBOT to fold them into this model. My plan moving forward is to focus training and updates on this model.
I do not need your buzz! If you like the model, please share some images and leave a positive review. Keep your buzz to help you create!
Disclaimer: This model is provided for educational purposes and for the advancement of AI art in general. The creator of this checkpoint assumes no responsibility or liability for the end user's outputs. Any illegal, unethical, or deep fake use of this model is the sole responsibility of the user engaged in those activities. The end user assumes all liabilities from use of the checkpoint.
No permissions are given for anyone to distribute this model on other sites. I do not give permission for it to be used on any image generation service other than CIVITAI.
Description
V2 - The V1 model was a complex model merge of 4 trained models with different core components (PONY, Illustrious, NOOBAI, and SDXL). V2 is training that model further then merging back with original mix to retain concepts. Pushing the realism conversion even further. It has solid NOOBAI prompt recognition.