This lora will not influence checkpoint style.
train data 752x752
0.6-1 for solo twins
more than 1 for 2-head solo
suggested checkpoint: Anything v3-v5，realistic
Negative prompt can be used a certain. Dont use negative prompt to fix bed images
DO NOT use "Solo","1","2","multiple" in prompt, this lora has no trigger words.
Restore faces is ok.
just support front-view.
This lora is experiment, please don't give me low stars
I have tried to train this lora for 5 times, it is hard to generate 2 head for one hundred percent. otherwise, overtrained.
这个lora我回炉练了4，5遍，过拟合的时候一定出现两个头，但是画面不能看，目前这个版本是最佳了，用1的权重，2-4 batch size 出两张以上双头。
Please give me 5 stars and like 😍, which is important to me.
🎉your reviews motivate me to update，your suggestion for updates are also helpful！
Hope any feedback from you, you can link me on discord, huggingface, github.
All train data collection are human-manual, and training cloud environment costs me much. hope your donation to support my new model. https://ko-fi.com/cecily34
If you have any questions related to the model or its samples, how to get started or just want to share your creations, feel free to join my discord: https://discord.gg/GgZFF3qPEB
That is my pleasure.
🌸 Loves U.
Description
FAQ
Comments (6)
It does not play nicely with other LORAs at all. If you include it with a weight of below about 0.6, it generates mostly empty backgrounds. Above 0.6, it does generate twins - but seems to ignore most of the prompts and almost all of the other LORAs you might have.
Yes, this lora will conflict with other Lora, my dataset is front-view, upper, twins (2 head). I guest other lora trained normal human image. have you tried some background LORAs or pure style LORAs? I am thinking and trying to improve train method. T T
Even without loras it won't follow my prompts at all. Always gives me the same people wearing the same thing, or just a background.
@skullzy77 will re-train it.
@Cecily_cc Awesome. It's a cool concept and probably not the easiest to train either. I'll play with it some more tonight using different models etc
@skullzy77 now, this model is training in my local pc. New version will be published within 24 hours.