<h1 id="heading-1465">About</h1><p>Another LoRA of a mostly obscure character but she's probably one of my all time favourites so I just had to. You can use any weight you'd like, I've had a lot of success just leaving it at 1.</p><p>LoRA contains trigger words for both her forms, Plutia and Iris Heart along with trigger words for their outfits. The outfit ones just make it easier for their outfits to appear since I had a lot of images of them in the dataset without their normal outfit, so make sure to use both if you want the outfit consistently.</p><p>I prefer version B personally but all of them generate slightly different images on the same seeds, download whichever you like the look of.</p><p>Preview image generation parameters <strong>should</strong> all be in the images as per usual (Feel like that should be standard but just my opinion). She looks sad or angry in all my previews because I forgot to put in an expression tag.</p><p>Finally, the generation parameters might look weird or different, I use the SD Next fork of A1111 created by Vladmandic, so that's why, it's pretty easy to figure out what the generation parameters should be for the images though.</p><h1 id="heading-1466">Training</h1><p>I did this one a little differently, I used the Prodigy optimizer in Kohya ss and used a decay weight of 0.4/0.6, there are 3 different models available to download, each is just a different epoch. I don't think there's a big quality difference between the three but look at the images and decide for yourself, all of them will be using the same seed and generation parameters.</p>

Files with SHA256: feb082aaaf42cb98172e1e40befd141e150fab778e9f44d8f09bcce9a9d47139