Model trained on NoobAI-XL V-Pred-1.0 to imitate the style of MoNo

=====

Support the artist: https://www.pixiv.net/en/users/93740696

=====

V2:

Overall more accurately reflects the artist's style and better captures the various concepts they usually work with. Quality tags will be very helpful in getting work that doesn't end up looking a little like sketches, but that aspect of the style is very much something I had wanted to capture as well. After training and testing 5+ different versions of the dataset to get it to work how I wanted, I am VERY happy with this version in the end.

- More robust dataset!

- Better tagging!

- No custom tags! (for example, got rid of pentransparent)

- Better censorship handling! (In V1, black censor bars had a tendency to show up regardless of tags, this version does a MUCH better job with censor control all-around)

=====

NOTE: All the images I generate for the showcase are done with base NoobAI-XL without the use of any quality tags, Adetailer, or Inpainting, only Hires. fix to do very basic upscaling.

My results do not look as good as you can get quite easily with more stylized checkpoints or better finetuning, but I wanted to show the baseline (and that the LORA works). Try out what works best for you!

=====

ORIGINAL V1 DESCRIPTION:

This was an absolutely massive dataset, so it is quite versatile. I would actually recommend a more finetuned model such as WAI-Shuffle (my personal favorite Noob-based model right now), as the style effect of this is strong enough to come through while still playing nicely with the more stylized models.

There are two tags that can help with certain things, but are inconsistent: "penshadow" and "pentransparent", mess around with them. If you've seen the artist's work it should (hopefully) be pretty apparent what these are for.

You may want/need to add "bar censor" and "breath" to the negatives, a LOT of their art has breath/musk clouds, so they show up frequently if not accounted for.