Type | |
Stats | 434 1,187 |
Reviews | (77) |
Published | Jul 3, 2024 |
Base Model | |
Usage Tips | Clip Skip: 2 |
Trigger Words | liru |
Hash | AutoV2 8A891C7B0C |
Liru || Magipoka
Pony: New dataset, better model, more Liru. Please enjoy this new version!
V2: Improved training, should be able to produce more consistent and more accurate looking Liru. Let me know how you like this version compared to V1, and I always appreciate seeing new Liru in the gallery~
V1: I've been sitting on this LoRA for a little while unsure if I should share, as it seemed difficult to generate Liru even after retraining 3 times. However, I've seemed to have found the golden combo of keywords that generate her pretty consistently, otherwise I think this LoRA can be used to make much better wolf girls in general. Used kohya_ss gui.
Pony version trained with Civitai On-Site Trainer.
Here are some parameters and helpful keywords:
ā¢ Trigger Word: liru
ā¢ Weight: 0.5-1 (Pony weight of 1 is fine)
Keywords:
V3: liru is sufficient enough to obtain accurate results.
V2: animal ears, tail, beltbra, shorts, thighhighs
Can add tanned or tanned skin.
V1: blonde hair, beltbra, dark skin, whisker markings
( try adding wolf girl, wolf ears, wolf tail if they aren't consistent )
Try using the LoRA without Liru's keywords and see how it improves your kemonomimi!
SD1.5: To produce similar results I recommend EasyNegative, this negative embedding was used in the sample images. I'm also using Clip Skip 2 and hires fix.
Pony: Images were generated with ComfyUI, check the image posts for workflows used.