Just an experiment with a dataset I used for SDXL training.

I use things "manga style drawn illustration black and white monochrome" in my prompt but it'll do color just fine, but I think the details are lost in some ways