trigger:
friendha,1girl,breasts,solo,mature female,white background,ring,jewelry,shirt,ponytail,mole,huge breasts,smile,looking at viewer,open mouth,simple background,wedding ring,jeans,hand on hip,pants,denim,white shirt,short sleeves,collarbone,hands on hips,
Description
FAQ
Comments (9)
Where is this character from?
Thanks, any chance you can share the data set?
@darkspaceinfinity sorry it lost in my disk, i have to find it in many dictionaries
@darkspaceinfinity very easy, just clip and duplicate them 8-10 images are enough
@lwander 请问剪辑+复制是什么意思？
@DSY666 Take what I say with a grain of salt, but I think he means get the one photo, crop it (face, upper body, lower body) into about 8-10 separate images then train it. This is for characters you want to make a functional lora with but have only one or a couple images. Results aren't as good as you would get with a diverse dataset obviously, but if you get good results with generation you can reintroduce that image into the dataset and train again, but with diminishing returns.
I remember seeing yhis one. Didn't thought it was possible to make a model out of it since only one picture of her exists... but im happy it does! Friends mother - more like, my personal plaything 😈
actually, there are many zero-shot technology like instantId and vibe in novelai, I just replicate it to 10 images dataset.