marcus-naslum/medieval_illumination_lora

Style LoRA trained on a 17 medieval (European and Persian) illuminations. Compatible with SDXL base model found on Huggingface.

The keyword that must be given in the prompt is: "a medu"

To use with ComfyUI, you need to download and use a conversion script on the LoRA layer keys. Something like:

wget https://raw.githubusercontent.com/huggingface/diffusers/main/scripts/convert_diffusers_sdxl_lora_to_webui.py

python convert_diffusers_sdxl_lora_to_webui.py \
    --input_lora data/medieval_LoRA/checkpoint-600/pytorch_lora_weights.safetensors \
    --output_lora data/medieval_LoRA/medieval_comfyui.safetensors

To use with Diffusers, something like:

import torch
from diffusers import DiffusionPipeline, AutoencoderKL

vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
pipe = DiffusionPipeline.from_pretrained(
    "/home/marcus-naslum/Documents/hf_models/sdxl-base-1.0",
    vae=vae,
    torch_dtype=torch.float16,
    variant="fp16",
    use_safetensors=True
)
pipe.load_lora_weights("data/medieval_LoRA/checkpoint-600")
_ = pipe.to("cuda")

prompt = "a medu of female beating priest, ferocious, colorful, snails" # @param

image = pipe(prompt=prompt, num_inference_steps=25).images[0]
image.resize((512, 512))