Description
FAQ
Comments (6)
"Excuse me, I have a quick question. Can you answer me, You’re training LoRA on pixan.al, right? Because I see the file is extremely large, 244.19MB, because only DIM 128 or 192 can produce 200-260BM file, and many characters don't have enough 30-80 images for training. I've also tried collecting and creating them, but it's still not enough images. So, do you gather enough images by using a dataset of AI-generated images? Can you share some tips.
i only use 10-20 images that i find online,mostly from google and zerochan and yes i use pixai.art
@broccolibaron 10–20 images for a LoRA is really crazy. I always thought you needed 30–80 images of every style to train a LoRA properly and get good results.
@lovemini Most characters aren't that complex 30-80 images is way overkill -- you can even train with https://modelviewer.lol
The only issue with training with images from in-game is it will be slightly more blurry than art
The most important part is to have good tagging / high quality samples that all look similar or you will run into issues
Hello, the prompt field is empty. Did you forget to upload your prompt?
I Think... RNDmissfortune,white hair,long hair,green eyes,glowing eyes,headwear,cleavage,one sleeve,one elbow glove,bare shoulders,dual weilding,pale skin