LoRA for Accelerator from Toaru Majutsu no Index.

Consists from 3 different outfits, which can be triggered by tokens provided below.

Recommended strength: 0.8-1. I personally find 0.9 to produce good results, as shown in preview pics.

Sampler: DPM++ 2M Karras.

Steps: 22-28 works fine.

CFG scale: 6-8.

Trigger words:

accel_ws - white shirt from TV-2

accel_bs - black shirt from TV-1

accel_bj - winter outfit

choker - adds choker

crutch - adds crutch

Suggested negative: (worst quality, low quality:1.4)

You can also add easynegative and verybadimagenegative_v1.3 if you want to, they do help with backgrounds.

If you have an issues with eyes there are two options:

1. hires fix / img2img sd upscale with 0.36 denoise and about 50 steps.

2. (Recommended) Self Attention Guidance plugin with 0.5-0.75 strength. (this will fix all issues without needing to upscale image, resulting in faster generations)

This LoRA was made https://civarchive.com/user/Simisir/, i just helped to him with dataset a bit and made some tests. Publishing it from his permission. Also please do check his another LoRA about Belfast from AzurLane here - https://civarchive.com/models/50516/belfast2-skins

Please share your creations as well, i will add best to preview.