!!! This is a Kolors (可图) lora. !!!

It is not an SDXL lora. Please use kolors models to use it.

You could download the weight from https://huggingface.co/Kwai-Kolors/Kolors

And use comfyui with MZ plugin https://github.com/MinusZoneAI/ComfyUI-Kolors-MZ/tree/main

This lora trained by https://github.com/lrzjason/T2ITrainer kolors slider lora training function.

Use prepare_slider_data to generate 10 pairs images for the training.

More config:

Main prompt: anime artwork of a beautiful girl

Positive prompt: highly detailed, well drawing, digital artwork, detailed background

Negative prompt: sketch, unfinised drawing, monochrome, simple background

Steps: 50

CFG: 3.5

Generation Batch: 10

Rank: 4

Repeats: 20

Optimizer: Prodigy

Lr_Scheduler: consine

Epochs: 5

Caption drop out: 0.1