LoRA is generated entirely from anime images.
-----------------------------------
Note:
I recommend against using 'asymmetrical sleeves', doesn't really do well on any checkpoint. Stick to 'short sleeves' or 'long sleeves' :)
Check samples for prompts.
-----------------------------------
Leave a like, comment, review but first of all, ENJOY ;)
Description
FAQ
Comments (4)
btw something i noticed is when you dont have a general keyword but instead use keywords for certain costumes in multi costume LORAs . When you use different costumes generations tend to be less similar to anime style. That is not an issue with LORAs without multiple costume keywords. Wonder if it is ebcause lack of a general keyword or because of multiple costumes making the LORA ""less burnt""
Overall general chara keyword is unnecessary and recommended to not use it as it causes outfit bleeds even when not used. At least according to the guys on discord who did plenty of tests on this.
i am not disputing that. If anything general keywords would probably ruin multiple outfit LORAs. Just pointing out that not having any keypoint at all in your prompt means artstyle is much less similar to anime than otherwise. Noticeable difference between with "nazaersuisu" and without "nazaersuisu". Just an observation.
then again, disclaimer: might be due to my checkpoint.
@Wizzi Either checkpoint or tags issue. If I want pure anime images, I only use 'score_7_up, anime screencap' and neg. 'realistic, semi-realistic'. Any other score in pos or neg prompt adds 2.5D style to the result.