这不能算是一个正式的版本,因为它不够稳定,达不到我的预期。
在过去的几天里我都在制作这个Lora,废弃了很多个版本,总是不能得到令我满意的结果,它实在是很难控制,跟AI稳定生成手一样困难。这个发布的版本只能说勉强可以使用,它不可避免得会出现错误的脚掌,以及多指或者少指的情况,我建议在权重0.6~0.8使用它,需要触发词foot up,需要主体词1girl或者其他角色Lora,否则可能只会生成脚。
最好是不要加姿势tag,可能会出现更加严重的肢体变形。
我建议可以多生成几张,把没有问题的放大,放大的重绘幅度最好是0.3。
建议只生成裸足,因为我还没有在学习库中加入丝袜。
我会继续优化这个Lora,直到它的错误率下降,请关注后续更新~
This is not an official release because it is not stable enough to meet my expectations.
In the past few days I have been working on this Lora, discarded many versions, always can not get satisfactory results, it is really difficult to control, just like the AI stable generation of hands. This released version can only be said to be barely usable, it will inevitably have wrong feet and the situation of too many or too few toes.
I suggest using it in the weight 0.6~0.8, which requires the trigger word foot up, the main word 1girl or other characters Lora, otherwise it may only generate feet.
It is best not to add a pose tag, which may cause more serious limb deformation.
I suggest that can generate more photos and enlarge the ones that have no problems, and the redrawing amplitude of amplification is preferably 0.3.
I recommend only generating bare feet because I haven't added stockings to my library yet.
I will continue to optimize this Lora until its error rate drops, please stay tuned for future updates ~
Description
❤️
FAQ
Comments (17)
真的有可能在sd1.5的总体框架下解决手脚的问题吗?这么久了都没有人能解决,感觉不是通过训练能解决的问题了
已经开始用穷举法排错了😵💫,手肯定不行,但是脚不是完全没可能,因为脚的结构和透视变化相比手要简单很多,尤其是这种固定它的姿势,没有最优解,但是应该是可以找出一个趋于稳定的值的,我感觉只要正确率有七八十都算可以了
是因为脚指头没办法玩出那么多花样是吗😏@Hayasaka_AI_
please do more feet content
5 toes on every image... it's a miracle?
use controlnet with depth and canny
@pantingyo как правильно его использовать ?
要白丝,要白丝!
Not having much luck with this lora. And it's heavily influencing the overall style of the gens towards a more traditional anime style.
I don't get foot fetishes, man.
why are you here?
现在ai绘图,其实就是在二维平面上把像素和颜色“正确排列”,没法从三维上出发,但是其实这也不是难事,只是还没有人把三维模型对接ai绘图这件事做成完整的功能,openpose是一个值得期待的方向,如果真的能从三维层面做ai绘图,那综合类设计专业才会真的享受到ai的福利,
需要丝袜版本!
需要丝袜版本!
关于foot focus, 如果可以的话,你可以研究裸足后 (加上鞋子+袜子 (指的是脱掉鞋袜))的Lora 模型





