Type | |
Stats | 284 4,430 1.9k |
Reviews | (48) |
Published | Aug 1, 2024 |
Base Model | |
Training | Steps: 1,920 Epochs: 20 |
Usage Tips | Clip Skip: 1 |
Trigger Words | illustration |
Hash | AutoV2 6D536818EB |
An art style LoRA based on illustrations found from Korean web novels, with a focus on historical fantasy manhwa and illustrative art (so lots of pretty big dresses and hats for women). Seems to do okay at higher weights like 0.95. Put "illustration" for extra impact.
As this was trained on images that almost all had blurry backgrounds or sparkly effects, backgrounds look like crap. I recommend "blurry background" or it'll look a bit strange and "light particles" "lens flare" in negative if you don't want that in your images. Men also look a bit weird sometimes? the proportions are very jank but it fits the style.
I also accidentally messed up the tagging for dark skin... but good news is I tagged all images as pale skin so just put "pale skin" in negative and "tan, dark skin, etc." in your positive prompt.
!!! Will maybe post different versions of this as they had interesting style variations !!!