Reckless (Lunatic) Emilia from Granado Espada
(Blue fonts are modified prompts)
emlu_(ge), 1girl, solo, mature female, large breasts, large ass, black hair, long hair, black eyes, dark purple makeup, purple lips, black ascot, white feathers, frills, white frills, frills between breasts, minimal clothes, revealing clothes, breasts apart, center opening, revealing back, lace, lace trim, detached sleeves, black dress, navel, feathers, black thong, micro panties, thighhighs, thigh strap,
claws, claws_(weapon)
magic, energy, dark energy, dark magic
Description
FAQ
Comments (4)
Hi, may I ask if you are using your LoRA together with any other LoRAs?
I tried some of your LoRAs, but I couldn’t get results similar to your sample images.
I use WAI-NSFW-illustrious SDXL checkpoint and Fine anime screencap LoRA only. Sorry for the confusion, I will tag them on suggested resources.
Nice start, but also several things wrong about her:
- She has green/blue eyes, not black
- She has purple lips, not bright red
- Makeup also purple
- Gem is purple, too
- Shes missing her white frills around her boobs
- Shoe details missing, lower cape details missing
- Her nose is bigger, but that could be tricky to fix (:
- Finger accessories missing, could be hard to fix, too :(
GE Mobile made her younger, not sure if her design changed much elsewhere there though
https://www.deviantart.com/michaelxgamingph/art/Reckless-Emilia-Granado-Espada-1001400007
Thanks for the feedback.
I somehow tried to fix them, but if the clothes are too complicated, the LoRA tends to not work as precise as original training data. My apologies for being noob.
- Purple lips and makeups could be fixed. It's my mistake omitting them, but when I made some sample data I preferred uploaded one so I just left it. I changed my prompt so please check it out.
- I didn't prefer finger accesories, but I also added prompts to add claw accesories.
Here are some problems,
- If the eye color is too complicated, it is difficult to make that exact color. Dark blue eyes, dark green eyes prompt tends to make more vivid color, so I just made it black.
- I tried to make frills on her breasts, but it just pops black frilled strings between her breasts.
- Since my training data is mostly upper body, (portrait 5~7, full body 5~7, upper body 20~30) if the leg details are complicated, it could make some poor results.
- Face anatomy like nose can vary depends on what checkpoints and LoRAs we use. It was tricky to make big nose in 2D AI artworks.
For training data, I only use official 3d artworks and ingame screenshots, about 40 to 50 datas. (Actually I also included your DeviantArt link image for my training data)No other fanarts since I'm afraid of breaching copyright problems.
Again, thanks for the detailed feedback, and sorry for the poor quality. I'm terrible newbie since I started training LoRAs only 2 months ago, and I have so much to improve.