竖着悬空捆绑_suspension、hanging_吊り｜LiblibAI
欢迎大家去lib在线生图返图和下载，毕竟他们是真的要给我小钱钱
本来是有横着和竖着的，但是做完之后，发现横着效果很不好，因为脚是折叠的，图跑出来很怪，就跑了一个只有竖着的，看着正常了许多，一些时候头上的绳子跑不出来。如果跑全身图的话，需要抽卡，一些时候，一些大模型无法体现出脚离地的那种感觉。
根据不同的大模型Lora权重建议0.5-0.9，有些大模型出不来这个动作，真人我没测
触发词是 suspension，建议加 bound, bdsm, bondage，可以加arms behind back，bound arms，适当加crotch rope，shibari over clothes。
2023.8.26
https://civarchive.com/user/VastoLorde
首先感谢这位老哥提供的训练集
Lora权重1，触发词suspension,feet off the ground，纸片人模型出图率比较高，绳索主要是在胸部的上下
需要全身都是绳索的话可以根据需求加 bound, bdsm, bondage,bound arms,crotch rope，但是需要抽卡
建议写个鞋子类型的提示词
用真人大模型需要抽卡，建议写一些人物正面的提示词，比如cleavage，navel之类的
半身图没测，只测了全身的长图
2023.10.18
首先还是感谢VastoLorde，在他的要求下发这个版本，他说这个比以前的都要好，具体过程可以看看我们的留言聊天
触发词是feet off the ground,suspension（the Trigger Words is feet off the ground,suspension）
负面提示词加ground,floor（the negative prompt you can add ground,floor）
需要全身都是绳索的话可以根据需求加 bound, bdsm, bondage,bound arms,crotch rope，但是需要抽卡
图片一定要长，例图是用的1080X2160；图片太长了，如果不点开看，是看不见头顶的绳子的
和其他人物lora混用，这个lora的权重可以0.6-0.7
https://afdian.net/a/TanedaRisa，网络乞丐，在线乞讨，非常感谢
Description
FAQ
Comments (24)
Doesn't work well for realistic models. This Lora is changing face for anime-like.
Thats true - not suitable for realistic pics
So, I have been experimenting with the new verson 2.0 for the past 2-3 days.
What can say is that it is working well, but has a few little flaws.
Some images seem to be lacking detail, though it has been varying on the Lora that was used or even small changes in prompts.
Another thing that the pose is often the same (character seen from side looking to the left.
Also in almost every case the ground is seen or better said, the feet almost touch the ground and I couldn't get a way to make the characters look like they are having high above the ground (this could and was in some ways varying from Lora to Lora.
Is there a possibility to get another version with more images than the 10 AI images used as this Lora's database.
I would even pay you for it :)
You can add A girl in the sky in prompt，and You can add ground and floor in negative prompt，Add weight to them。Generate ten images，One or two images have this effect。If you are not satisfied, you can send me more images，I will try again。ps.I have uploaded the new image, you can take a look
@TanedaRisa Yeah, it seems to be working better. I would be really interested how the Lora would turn out if you use more than the 10 AI images you have used as database. I have sent you quite a few pictures for reference. I have more, but I don't think they would help. I have already chosen the best for it.
@TanedaRisa It also seems to have problems generating detailed images when using it in combination with a character Lora or other Lora in general, without it it is high quality.
@VastoLorde But the images you gave me，I can only choose ten high-quality ones，The remaining images are not suitable for making Lora
@VastoLorde In the future, I will collect more images and try to create a better one
@TanedaRisa I can look for some more later
I have sent you two seperate mails with higher quality images, these should definitely help :)
@VastoLorde Thank you very much，I will try again
@TanedaRisa Thanks a lot! ^^
@VastoLorde I will send you an email，There are Lora models and xyz images inside，
@TanedaRisa Thank you, I will look over them later! :)
@TanedaRisa I have tried them and they seem to work better a lot. The ground is often not visible and the feet are not touching the ground anymore.
How many of my images did you use for the database? Also what is the difference between the versions 000001 to 000009?
I just added them all to my Stable Diffusion :D
@VastoLorde I used approximately 20 of your images.The difference is that the number of training steps varies from few to many.the xyz image,You can see the subtle differences.The proportion of AI images used this time is relatively high,So, sometimes the body parts of characters may have defects.
@TanedaRisa Ok thanks!
@TanedaRisa also you could upload the other versions here as well, the seem to work better :)
@VastoLorde I think the current version is good enough，This new version is only for you
@TanedaRisa Ok thanks. Can i do anything in return?
One additional thing it took me 2 months to find out is that if you use a character Lora in combination with the suspension Lora, reduce the weight of the suspension Lora below 1, like 0.6-0.7.
Otherwise there are some issues with the character's clothing or other aspects.
At least when using the models 000005 to 000009 you gave me. Maybe you should upload them as well. They are working quite a bit better than the versions here.
ok,I will upload it in the future
you think the 000005 to 000009,who is the best,I will choose the best upload you think
@TanedaRisa I have been using 000009 mainly, have to check which works best.
@TanedaRisa lora:feet+off+the+ground,suspension-000009 seems to be the best when it comes to combining it with a character Lora