Description:This is my first LoRA, and it's entirely possible that I have no idea what I'm doing. Noticed that there weren't any LoRAs for this spunky lynx chavette, so I figured it'd be a good first. Of course, there was also very little in the way of official or fan reference material, so all of the training images were made by me. Tried to stay as on-model as best I could, but I could only do so much with this first version. The next version(s), when I get around to making them, should be closer.The clothed tag will more-or-less get her dressed in her normal outfit. If you leave it off, you'll probably get something similar, though the style or colors will almost certainly be different. The nude tag will remove all, or most (she likes to keep her boots on sometimes), of her clothes. While the training data had lots of examples of her nose piercing and eyeshadow, it doesn't always generate. Adding in those tags will pretty much ensure they show up. It also likes to give her piercings in other areas as well, though. Especially when combined with the nude tag. Lastly, I'm not super happy with her nose from the training dataset, so adding in small_cat_nose gives better on-model results.This model handles both SFW and NSFW equally well. Seems to play nicely with other LoRAs as well as Textual Inversions (<a target="_blank" rel="ugc" href="https://civarchive.com/models/87781/boringe621-negative-embedding-enhance-images-stylistically-and-topically">Boring_E621</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/7808/easynegative">EasyNegative</a> work especially well, but be aware that they will portray her as being much younger than she actually is).Suggested Tags:Positive: lynx, 1girl, nose_ring, small_cat_nose, eyeshadow, clothed, nudeNegative: young, child, cubTraining information:Model: fluffyrock_3mE86OffsetNoiseE71Training Dataset: 49 images of my own creationRegularization: 38 images of fan work and movie stillsEpochs: 16Steps: 4500Resolution: 786x786Known Issues:Since this was trained on FluffyRock, it will work best with models derived from that and other models are likely to produce poor quality material.Since this first version was trained only on data I made, the LoRA may be pretty rigid. I did my best to include different poses and emotions, but I'm only one man. Future versions will ideally be made with works from other creators that use this.All of training data had sterile backgrounds, so if you don't specify any, that's probably what you'll get. This isn't really an issue so much that it's a tip for generating better images.

Files with SHA256: a336be627d8df83cdeffb986a119bd4d90652c6122fceddc2ad6df8876df6b52