This was trained purely on screencaps of the monogatari series (mostly Bakemonogatari so far).
I tried to capture the general visual mood of the shows as well by including a bunch of the backgrounds. It worked out semi-decently.
Updated
Characters in general should work way better now, because of the added data.
The shading on squishy bits is more like in the show as well now.
The outlines are more pronounced and colored properly like in the show most of the time now.
If you want to support me buy me a coffee!
Weight: 0.8-1
Triggerwords: None.
Notes:
Sometimes produces closeups when not prompted, just put close-up in the negative prompt.
The backgrounds tend to work out better when generating pictures in landscape/horizontal resolutions rather than portrait/vertical ones.
It will sometimes ignore solo/1girl/1boy and proceed to create a landscape-shot, just add no humans to the negative prompt.
Prompting from behind will make the LoRA almost always focus on the subject's butt for better or for worse. To prevent this just put ass focus and/or ass in the negative prompt.
Feel free to share some of the pictures you create with me and post them here!
Description
Added a lot more images to the existing dataset and removed some that were too blurry/unfocused.
Also trained it at higher network dim , which will hopefully result in more details being absorbed.
(Sorry for the filesize increase)
FAQ
Comments (9)
1. the fact that It can pull the iconic head turn by itself is genuinely impressive, will try later and left some stars.
2. As per usual in any style LoRA/Lycoris/LoHA that doesn't include her, I must say: No Megumeme, 0/100
I am actually a Megumin-hater, Darkness all the way. Sorry to disappoint! x)
Also the headturn is still a gamble to pull off!
So don't get your hopes up too high, but I already updated it and fixed some things that bothered me.
Wow. I have been waiting for a LoRA for this and it turned out better then I thought it would.
what does the NUM vale do in the loras your the only person I've seen do it?
Oh the NUM value is just what I use for making the testing image, in this case the third one.
https://files.catbox.moe/8loy6q.png
I usually do this for the testimage and it just generates an image and replaces the NUM with the values you enter, so it just raises the LoRAs weight in this case. It's pretty handy as a visual guide. You don't have to call it NUM either I think, it's just a variable!
卧槽，这才是绝配我有小忍模型
嘿，这是自动翻译的，因为我不会说中文 x)
如果您喜欢，请随意发布该模特的图片并留下评论！
Been loving this. Captures the desired vibe of characters, lighting and background art pretty consistently to the series' style while being flexible with other LoRAs and checkpoints. Thanks!
Possibly one of the greatest LoRAs to ever exist. You have done a great service for humanity, sir.