10 months later, I've finally updated this thing. And hoo boy did it need updating! V1 was my first LORA, and it showed. V2's tagging is more consistent, it is trained on NAI instead of AnythingV3, plays much nicer with style LORAs, and doesn't melt her eyeballs anymore. The dataset is also 4x larger and features more varied angles, NSFW, and alternate outfits. You no longer have to prompt for orange hair, ahoge, or green eyes (usually), and can add more weight to "curly hair" if the model you're using doesn't understand her cinnamon bun curls.
V2 is trained at 1024x1024, which is high for an SD1.5 model and can sometimes cause issues with highres fix, but seems to preserve details better. I've also uploaded a 768x768 version for those who prefer it.
Some of the preview images aren't one-shot txt2img, but the metadata should give you the base I worked from.
