Have you ever wanted to see white or "occidental" women in impractically tiny Japanese cut out swimsuit lingerie? Or thicc women, women of color, or tatted women doing such a thing? While PredatorRat is doing the lords work dressing their flawless hyper-confident lingerie models in impractical non-clothing that's too small to move in, one must admit they have a type, so to speak. And this is where I come in.After much grueling quantitative research, wasted runpod instances, and stupidly accidentally overwriting a version that actually worked, I've taught AI how to make Japanese cut out swimsuit lingerie, I know, I know, thank you, gentlemen, please hold your applause, you're all too kind. With the power of AI, you too, can jam, I dunno, whoever the fuck?, how about "your mum"? into some Predator Rat t-back spats. Can I get a hooah?Trigger words are <code>spa t-back spats</code> you can throw in <code>blue</code> , <code>striped,</code> and <code>raglan shirt</code> if you want (the rationale was <code>spa-0032</code> is the product number for this particular item, I may retrain using this word with a more advanced captioning model such as "moondream2"). Maybe someday some one can tell me why prompting InstructBlip to include <code>t-back spats</code> into the captions made this work when the most precise and intricate descriptions of what it is actually in each picture failed.

Shimimaru×PredatorRat「T-back Spats with Line」〈BK×WT Line〉 紙魚丸×PredatorRat「Tバックスパッツ ライン入り」〈黒×白ライン〉

Trained on the same data and with the same settings as the 1.5sd lora. Doesn't seem to have the same quality, however, I suck at prompting sdxl models and have almost no assets for them. Let me know what works.

Files with SHA256: 0029e64707db3814508ea34e4e3bd26d40441e20c24cee41082d2b22bc887724