(Illustrious/Pony/1.5) Elh Melizée (Solatorobo: Red the Hunter)

Elh Melizée from the Nintendo DS action RPG "Solatorobo: Red the Hunter." The 2nd game from the "Little Tail Bronx" series from CyberConnect2.

Trigger word: elh

Useful Tags: 1girl, furry female, cat ears, red eyes, orange eyes, short hair, white hair, facial mark, animal nose, striped tail,

Outfit: elhattire, earrings, fur trim, striped sleeves, white boots, black pants

Other optional tags to mess with: shiny hair, body fur, fur-trimmed boots, fur-trimmed sleeves

Optional but potentially useful Negatives:

striped fur (For when her fur has random purple stripes/markings. Especially common on her arms when not in her normal attire. Nothing in the dataset has this and I think her striped sleeves and facial mark partially confuse it.)

colored tips (For the.... shading? that is almost always in her hair towards the back. For if you want it to be one solid color. Might need increased weight to fully work consistently.)

Since both the game's official art and fanart cannot decide or agree on some of her details, you can change any details to whatever you want. Much of the dataset has her eyes yellow or her hair grey, so you can use "yellow eyes" or "grey hair" instead if you'd prefer.

Don't even get me started on the X on her chest... which is also on her back in some of the fanart. It will be on her outfit sometimes, and be completely gone other times... Which is accurate to the game so I'll consider it a feature :v)

Illustrious V1:

Wooooo, another excuse to make a new version!!!

First attempt at making an Illus LoRA. I just reused the exact dataset from Pony to test it out and... I think it turned out pretty good. Like the results of this over Pony almost every time. Same promptings and whatnot as Pony.

Try out different weights. Mostly used 1, but lower weights may work better in some situations. Her outfit was being weird from behind and lowering the weight seemed to fix it in that one situation.

Pony V1:

Cut down and heavily re-modified the dataset to 50 images. Fixed a few more captioning mistakes on the remaining images.

SD1.5 V2:

55 New images added, almost all being heavily modified images from the official fan artbooks. 142 images total.

Fixed up some bad captioning and added more. I also touched up some of the images from V1.

For this version only, use "strap" in the prompt for a better outfit. Uses same prompts above otherwise

Trained on Indigo V105, 512x768 at 5680 steps. Gives good results with a few select models, but still struggles with others.

SD1.5 V1:

My first LoRA ever.

Trained with 87 images at 7656 steps.

Trained on neatnessFluffyFurMix_zephyr and trained at 512x512.

Solid results on most kemono/furry models, but inaccurate on others. On some models, may give better results over V2

Uses same promptings listed above.