v1: Trained on a few episodes from the first OVA, only one outfit.
v2: Trained on every episode from the first OVA plus the special episode, three outfits, and a more consistent dataset redone from scratch.
Many of the images had dark lighting, so some images may come out looking darker. Adding "night" to the negative prompt can mitigate that, though the colors may look a little too saturated.
ADetailer (or FaceDetailer for ComfyUI) is highly recommended. Annoyingly though it tends to add black lines to her face even when they don't make sense, adding "blush" to the negative prompt sometimes helps with that.
https://civitai.com/models/1032344/ryoko-hakubi-tenchi-muyo-ryo-ohki-illustrious-xl
