以前公開したterasu mc風hyper networkのlora版です。
個人的には結構満足の出来です。ただ漫画の一部を学習に使ったので白黒画像や文章がたまに出てきます。気休め程度ですがネガティブにmanga、doujinshi、textを入れています。
480枚の画像で学習したみたいですが、作ったの結構昔なのでどうやって作ったかは忘れました。知りたい人はメタデータを見てください。
This is a lora version of the terasu mc style hyper network that I previously published.
Personally, I am quite satisfied with the results. However, since I used a part of manga for learning, monochrome images and texts sometimes appear. I have included manga, doujinshi, and text in the negatives just to ease my mind.
It was made so long ago that I forgot how it was made. If you want to know how I did it, please look at the metadata.
(deepl translate)
Description
FAQ
Comments (3)
that moment when your translation tool is active and you wonder why there is 2x the same text... facepalm
wondered how many photos did u use to train?
I looked for the dataset and found it.
It looks like it was trained on 480 images.