Edit (20230529): I was playing with this LORA and it is sadly showing it's age. though it is only 2 months old it seems an eternity ago. Regardless went and fixed the tagging to make it produce some of Ranma's iconic outfits.
Here are the new triggers of the outfits:
RanmaRedShirt (normal outfit)
RanmaUnderwear (boxers and tanktop should be able to produce her orange tanktop variation)
RanmaLeotard (uses both green and pink in the dataset should be able to produce both but it seems to prefer green)
RanmaJungleDress (I don't remember where this one comes from just a tropical looking dress)
ovadress (her dress and alternative hair from the ova)
Had good results with: RanmaChan, single_braid, large_breasts, <lora:RanmaChan:.7>
Remember to add twin_braids in the negatives and use the common positive and negative tags and embeddings.
**************************************************************************************************
I had already collected the dataset but I was beaten to the punch, so give them a visit too. That Ranma LORA is nice but is a bit softer than i like. Then again this is a ???? version made mostly with fanart and for gentle-beings of culture.
It ended up mainly a mixture of Mage, with a few sprinkles of Mogudan and Naruhodou and a boatload of crappy 90s shit. So YMMV.
Using chinese_clothes seems to give her the dress version of her clothes while using tangzhuang seems to give her pants.
Using single_braid might be a good idea as the LORA sometimes gives her twin pigtails. Also put red_hair and blue_eyes, as those can drift. I don't think i put too much black and white stuff yet it still happens.
Seems to work best from the .35 to .8 range. Best to keep it in the middle so .5 or .55.
Had good results with: masterpiece, best_quality, highres, RanmaChan, red_hair, (single_braid), blue_eyes, ???????????, <lora:RanmaChan:.5>,
The typical good prompts seems to help with some of the graininess of the dataset. Then again 90s lowres shit. I tried to do some up scaling to mixed results.
I added the tag summary to the dataset if you wish to experiment with outfits and "poses".
Seems a bit overcooked and I might have done something iffy with the triggers by reordering them strangely. Anyway the trigger RanmaChan needs either 1girl or something ranma-ish to properly trigger. I was surprised when the first thing that appeared was a male Ranma looking dude, which i am sure came from the faceless randos + Ranma herself from the ??? images in the dataset.
