No.
1: LK2 in particular is a game I have fond memories of, while I find LK1 its less refined predecessor (I actually played LK2 first).
2: There's less official 2D art of her, and LK1's model is way more primitive
3: I do not want to tag stuff again for a while after the giant task that was hand tagging the Gundam X style LoRA (and when I am up to it, the first thing I'm doing is doing a 2.0 of my Mana LoRA since I failed to include a big chunk of good training data)

My advice if you do decide to try, assuming you've got a good PC 1: Use Dolphin (after you've ripped your Lost Kingdoms 2 disc with a modded Wii of course...) and use its freecam function to get good shots in multiplayer (since it's better quality than her LK1 model). Also rip her model and look at it in a model viewer/editor program (I'd recommend having a low level of texture filtering on for both of these) 2: Tag "3D" and "game screenshot". This helps minimize the style influence just using a bunch of pictures of models will have (especially with 3D in negatives). 3: If you're looking for fan art, Katia's Japanese name is "カティア" and the series is RUNE or ルーン (you will need to search using both because neither is unique). I don't think there's much out there, but Pixiv has a few pieces and you might find some more on Twitter. 4: Here is a picture of the thumbnails of my training data to give you an idea of what I gathered beyond the raw numbers ( I also wouldn't do pics from the left and right for Katia since her design isn't asymmetrical as far as I know. In retrospect I'd try to include more variety in the backgrounds of the pictures and have taken 24/25/26 in separate locations for each)

@Sargovel All my stuff is trained on site. The good computer part was in reference to running Dolphin, which is trivial compared to even SD 1.5 (I think even modern CPU graphics can run it on lower settings at this point). Only changes from default on trainer I've made were enabling "shuffle captions" on everything, and changing clip skip to 2 for the Asticassia uniform and Gundam X Style LoRA per tooltip recomendation (as both were 100% anime screenshots). Texture filtering is a simple option in Dolphin or Blender.

As for the tagging, I've never done anything particularly complex. Just figure out what you're going to tag a character with ahead of time (like what tags make up their outfit, hairstyle etc.), and apply any applicable tags (e.g., don't tag "shoes" if only upper body is visible) to the image using consistent tags (as well as any fitting tags for the background when you can) while removing any incorrect autotag results (so you don't have images tagged "1girl, 2girls, solo, 1boy"). For example, picture 24 in the linked image is
"3d, game screenshot, tara_grimface, 1girl, solo, from behind, blue nails, facing away, standing, blonde hair, high ponytail, short ponytail, blue hair tie, hair tie, single earring, necklace, crop top, blue crop top, sleeveless, bare shoulders, black capelet, red trim capelet, capelet, asymmetrical arms, arm wrap, nail polish, blue nail polish, shorts, blue shorts, belt, runestone, boots, black footwear"
(I'll provide any pic's tag set if you ask)

@Sargovel I think the trick is just having pics in the data set that won't have a particular part visible so it learns what they are by their absence.

@Sargovel Yes. Already scheduled for the 10th of next month. Everything except moon (didn't work as well as I'd like and someone else did it better. If anyone really wants it they can have the tagged training data since it's all public domain pictures) and Djinn (didn't come out to well, a V2 is on the eventually list) is getting converted to IL.

@Sargovel Up now.