LoRA is designed for use with ponyDiffusionV6XL or mix, for example AutismMix SDXL

I made separate versions for the demon clothing and for the school uniform.

There is a problem with the pupils. The ponyDiffusionV6XL model does not know the +_+ and symbol-shaped_pupils tags, so knowledge about this comes only from my LoRA.

There were a lot of problems with the school clothes version. I spent 2 days and 8 attempts to improve the result. But I couldn’t do anything better, except not to combine two types of clothes in one LoRA. Most likely we need more good images.

Model: ponyDiffusionV6XL_v6.safetensors

Dataset size: 64 demon clothes \ 18 school uniform

Clip skip: 1

dim32, alpha16

lr=0.0003, constant, adafactor - demon clothes

te=0.00025, unet_lr= 0.0005, pony_sdxl_vae, constant, adafactor - school uniform

Optimizer extra arguments: scale_parameter=False relative_step=False warmup_init=False

My tips for generating on SDXL for AUTOMATIC1111

Resolution:

• 1024 x 1024

• 1152 x 896 or 896 x 1152

• 1216 x 832 or 832 x 1216

Don't use Hires. fix. Send the picture to img2img, set CFG 0.15-0.25, then select “SD upscale” in the script, you also need to select 4x-AnimeSharp (find out how to install it) or what you like best from upscale. All other settings are default. Click generate.

For me 12gb VRAM is enough for this using --xformers