
A very powerful LoRA trained to recreate many styles and concepts, based on League of Legends splash art.
This is not meant to just be a "set it and forget" it LoRA, you need to use it properly. Read below.
I recommend starting with this default prompt:
POSITIVE
l0lspl4sh, source_anime, score_9, score_8_up, score_7_up, year2024, year2023, West Studio, sixmorevodka, {prompt}
NEGATIVE
score_4, score_5, year2009, year2010, year2011, year2012, year2013, year2014, crazy eyes, constricted pupils, legacy, {prompt}
This LoRA really loves highres fix, as it brings out a ton of detail. Additionally, i can only recommend using Adetailer, to get perfect faces.
I recommend using these parameters:
The recommended prompt "template" is as follows:
l0lspl4sh, source_anime, score_9, score_8_up, score_7_up, <yearXXXX>, <CHARACTER|1girl/1boy>, <SKIN NAME 1girl>, <SKIN SERIES>, <Artist1>, <Artist2>, <general tags>
The images were tagged exactly as they appear on the league of legends wiki:So if you're going for something specific, maybe look up skins for inspiration from the wiki.
The model was trained on 3000~ appropriately tagged images, with many new concepts introduced. Take a look at the wordclouds below to see which tags are most prominent in the training set. (the bigger the words, the more images had the tags)
Artist Name
Release Date
Skin Name
Skin Series
描述:
"Initial" version. Many iterations in, the first I think is ready. Unlikely to be updated.
训练词语: L0LSPL4SH, year2024, year2023, West Studio, sixmorevodka
名称: L0LSPL4SH_40epochs.safetensors
大小 (KB): 752042
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success