This was trained on the chibi artist xinzoruo. It generates both backgrounds and characters as you can see by the preview images.
I would recommend using it with AOM2 at weight 1 as for prompting I would recommend putting "yelan (genshin impact)" into the negative prompt as she is over represented in the dataset and most if not all characters will come out looking like her. Adding her name to the negative prompt evens it out a bit although she will still show up more frequently than not.
A non character focused version is planned so I can improve the consistency and quality of characterless images.
Trained on AOM2 using some random settings from a youtube video
I still have no idea what im doing