Houraiji from Houkai Gakuen 2I had a not so good prototype LoRA I made of her long time ago lying around my drive for quite a long time, so I decided to re-train her with a better dataset. Which was quite painful to build considering the lack of high definition images and the lack good quality fan art, ripping the game resources helped a lot however.Her dataset still has room to improvement so you can expect another version at some point.Tested with a weight of 0.7 and 0.8 (check pictures for prompts).Her main trigger word is houraiji, I can confirm changing her clothes is possible but lowering her weight might help sometimes, using the helper secondary trigger word houraiji classic alongside her main trigger word should reinforce her main clothes appearance.Some additional tags for default outfit: serafuku, purple hair, two tone hair, pantyhose, midriff, bandages.Negative Prompts: (worst quality, bad quality), (easynegative:1), ((negative_hand-neg)), low resolution, fat, thicc, fused legs, distorted fingers, ((lips)), text, signatureTrained (mostly) on 768x768 images, it will only perform well at high resolutions with Hires fix (Settings I use: 4x-UltraSharp upscaler, 10 hires steps and 0.4 denoising, upscaled by 2.4 from 648x800)LoRA trained with NAI, images rendered with FoxyaNSFW V4 using the klF8Anime2 Vae.NOTE: I will try to cover as many characters as possible from this old godforsaken game as long as:1. I like the character enough and I can find decent enough quantity of good quality images to work on an usable dataset (sometimes when I like a character enough, I make the fan art and concepts myself to feed the datasets but since I'm not an AI, it takes a lot of time).2. There isn't enough LoRA of the character in question already (Many characters from Houkai Gakuen 2 already appear in Honkai Impact 3 and many have enough good quality LoRA already).

Houraiji Kyuushou | Houkai Gakuen 2 (Guns Girl Z)

This is actually a re-training and improvement of an old prototype I had, still has room for more improvement.

Files with SHA256: 1f80cbe05c8cd6b304abb1f10aec95cba7d63370dcf85dfaab3e0920c2211554