Most "stained glass" models (if not all...) tend to create a stained glass background and then put the character in front of it, without really blending the character in the stained glass. I created a handmade small dataset of images as close as possible to what I really want from a stained glass style. V1 of this LORA was created with that dataset.

This LORA will be used to create a more extensive dataset for a V2, but my local tests seemed good enough for me to share the model, so here you go.