[PonyXL B-LoRA] Nao Miyama 美山ナオ | 365 Days to the Wedding | Kekkon suru tte, Hontou desu ka 結婚するって、本当ですか

Nao Miyama
from
365 Days to the Wedding
(Kekkon suru tte, Hontou desu ka 結婚するって、本当ですか)

This model was trained on anime screencaps (up to Episode 7) and two AI generated images.

Note:
This is a B-LoRA. Use the style LoRA to adjust the intensity of the anime style.

Prompts:-

• Trigger Words:

nao_olis, tan, 1girl, brown hair, brown eyes

Hair Styles:
long hair
high ponytail

• Outfit 1:

red tank top, white shorts

• Outfit 2:

t-shirt, black pants

Negative Prompt (you may need this to prevent outfit bleeding):
white shirt

• Outfit 3:

white shirt, cargo pants

• Negative Prompts (optional):

backpack, shoulder bag
blurry background, brick wall

Weights:

Character LoRA: 1
Style LoRA:     >=0   (0 is fine if you don't care about her outfits or the anime style)
                >=~5  for a more accurate anime style
                ~1    for a more accurate outfit

(Feel free to experiment with the weights.)

Known Issues:

• You may need to add backpack, shoulder bag to the negative prompt when using t-shirt or white shirt in the positive prompt.

• t-shirt may turn white. Adding white shirt to the negative prompt usually resolves this issue.

Afterword:
I continued experimenting with eye inpainting. While checking out models created by others, I noticed that some experienced trainers use ADetailer with face YOLO models for automatic face inpainting in their samples. I then decided to use a script generated by ChatGPT to automatically crop portrait images in my dataset using face YOLO model, speeding up my workflow. During inpainting, I simply inpaint her eyes using ADetailer with the same model. However, those highlights in her eyes (not sure what they're called) in this model are a bit inconsistent. For my next models, I might work on ensuring more consistent eye details, either by removing problematic images from the dataset or modifying them myself.

Hopefully some of you will like this model.