I have a form for requests; check it out if you want to request a character.
Mugino Shizuri from A Certain Magical Index and A Certain Scientific Railgun. Trained on Illustrious-XL v0.1. Tested at 0.8-1.0 strength.
Contains her Railgun outfit and her Index outfit.
Trigger word is "mgszr" and "brown hair, wavy hair, long hair, brown eyes" is required.
Add "purple dress, long sleeves, cleavage, black belt, short dress, black thighhighs, white boots, high heel boots" for her Railgun look.
Add "checkered scarf, beige coat, layered sleeves, yellow sleeves, brown belt, yellow pantyhose, high heels" for her Index look.
A request from the form. Check out my Discord if you want to chat about video games (or yell at me about requests.)
Description
FAQ
Comments (3)
Hi there!
May I ask something about your training setup?
I really like the quality of your 54 MB LoRA — could you share what training parameters you used (epochs, rank, etc.)?
I'm currently training my own character LoRAs with this setup:
UNet LR: 0.0006
Rank/Alpha: 16 / 8
Scheduler: cosine_with_restarts (3 cycles)
Optimizer: adamw_8bit
Steps: ~4000 (5 epochs)
Noise offset: 0.07
Text Encoder LRs: 5e-5 (both encoders + embeddings)
Preserve embedding norm: enabled
The results are solid, but I’m curious to compare approaches with experienced trainers like you.
Thanks in advance and great work on your models!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.