I RECOMMEND WORKING WITH A MINIMUM RESOLUTION OF WIDTH 512 X 1024 Heitgh to get the results i get in the published examples.

I AM NOT GOING TO UPDATE THIS LORA

If you want to improve or create a lora with the attached dataset, I recommend first reviewing the data set, since it has images of male characters which could ruin your training, I recommend only training female characters SOLOLY, only one female character per image

READ ALL VERY IMPORTANT

I recommend use the generated imagnes in this post as examples of how to use this lorA.

BETA LorA trained to follow the style of 3d artist Moiarte, you all could find the artist and help his content in Twitter Moiarte3D

KEYWORD PROMPT (use): Moiarte3dstyle

I recommend using this way the lorA: <lora:Moiarte3dstyle-04:0.7>, Moiarte3dstyle

put 04 in epochs and 0.7 as weight

Checkpoint: ponyDiffusionV6XL

Base model: Pony Diffusion V6 XL

To get decent results (use):

-Recommend:

• sampling method: DPM++2M (BETTER) or Euler A

• sampling steps: 120 or more

  -everything else by default

• hire fix:

  >Minimun Hire steps: use 10 less units or the same amount as sampling methods

  -everything else by default

• refiner: default

• CFG: scale: 6.5 or 7

The tool used to tag the images is: https://colab.research.google.com/github/hollowstrawberry/kohya-colab/blob/main/Dataset_Maker.ipynb#scrollTo=sl4FD7Mz-uea

The model can improve especially if the tags generated by the tool are edited, since the tags provided by the IA tool ARE VERY GENERAL and not very specific on different issues. That is the reason why the model needs many prompts to obtain results close to what you requested.

I have attached the data set maker if you want to do it yourself and improve/create a new lora with this style (READ VERY IMPORTANT: THIS DATA SET ALREADY HAS MALE CHARACTERS INCLUDED, FOR THIS NEWLY RELEASED LORA MODEL I DID NOT USE MALE CHARACTERS.

I ONLY USED DIFFERENT FEMALE CHARACTERS, PLEASE NOTE). THANK YOU