A style LoRA for Moss Image CFT trained on 200 images from the ByvanckB dataset for 1800 steps.
I'm impressed by how well the model has learned the style to be honest.
The model works even without it, but for best result, remember to prompt the fact that you want a medieval illumination.