Sign In

[PonyXL] Kurotaki Rin 黒滝凛 | Saishuu Chikan Densha

49
276
6
6
Updated: Aug 7, 2024
characteranimewoman
Verified:
SafeTensor
Type
LoRA
Stats
230
6
Reviews
Published
Jun 3, 2024
Base Model
Pony
Usage Tips
Strength: 1
Trigger Words
rin_olis
Hash
AutoV2
8A8E9D4D50
default creator card background decoration
JasonCMK's Avatar
JasonCMK

Kurotaki Rin from Saishuu Chikan Densha

This model was trained on both anime screencaps and game CGs.


Changelog (v1b):

  • 1boy should work more consistently now.
    (This may seem odd since I uploaded both versions simultaneously. However, I only noticed this issue after generating most of my sample images, and I am too lazy to re-generate them using the new version. So, please check out v1 for the sample images.)

Base Model: Pony Diffusion V6 XL (https://civitai.com/models/257749?modelVersionId=290640)

Prompts:-

  • Trigger Words:

rin_olis, black hair, long hair, purple eyes
  • Outfit 1:

peaked cap, black suit, armband, pleated skirt, red thighhighs
  • Outfit 2:

peaked cap, black jacket, open shirt, black necktie, armband, skirt lift, red thighhighs

Optional:
purple bra
purple panties
bra lift, breasts out, nipples
shibari
  • Styles:

anime coloring (anime style)
game cg (game cg style)
  • Optional:

1girl / solo / uncensored
  • Negative Prompts:

grey eyes
censored, mosaic censoring, bar censor

Known Issues:

  • Her jacket may merge with her shirt or become an open jacket, exposing her shoulders.

  • As usual, asymmetric items like armband may appear on the wrong side or on both sides.

Weight: 1 (feel free to lower it, especially if you don't care about her outfit)


Afterword:
I just realized that OneTrainer allows a token limit of 75 only. This is fine for normal characters, but for hentai characters, I had to spend more time pruning tags. Additionally, I have been trying to train my SDXL Lora with Weight Decay=0.5 (on Prodigy with d_coef=1) for my two recent Loras. The results seem to be better than what I did in the past, where I trained several models with lowered d_coef and then picked the best one.

Hopefully some of you will like this model.