『トラスとウェッジ』は女の子アバター「トラス」とフクロウ（？）のアバター「ウェッジ」で学習したLoRAです： https://nylonheart.booth.pm/items/3472207
良いweightは0.6-0.8ぐらいかなー
これはanything-v4.5-prunedで学習させたLoRAですが、他のアニメモデルでもうまくいくはずと思います。
コメントやフィードバックは大歓迎です！
=============================
Trained LoRA on the female 3D Avatar named Tolass and her companion owl (?) Wedge: https://nylonheart.booth.pm/items/3472207
Good weights is around 0.6-0.8.
This was trained with with anything-v4.5-pruned but should work well with other anime models too.
Comments and feedback are welcome!
Description
ヨネマルさんのTwitterやBoothからのトラスちゃんの写真とスクショ120枚を使っていただいて学習させた。
Trained using 120 images (12000 steps) from the author's Twitter and Booth pages.
FAQ
Comments (3)
Hi, this look very good!
Would you mind sharing your workflow.
What kind of pictures you took in vrchat?
Did this require chroma/greenscreen world.
Best of luck! ✺◟(＾∇＾)◞✺
-
machine translated:
Kon'nichiwa, kore wa totemo yoidesu ne! Wākufurō o kyōyū shite itadakemasen ka. Vrchat de don'na shashin o torimashita ka? Kore ni wa kuroma/ gurīnsukurīn no sekai ga hitsuyōdeshita ka? Ganbattekudasai! ✺◟(^∇^)◞✺
Hi, thank you!
Sorry for the really late reply, I have been inactive.
I didn't take any pictures as I only used images from the avatar's author's Twitter and Booth pages. It didn't matter if the pictures had background or greenscreen, I just used them as it is. Cropped the images into 1:1. I used 120 images in total and trained with 100 steps (12000 in total). I mostly based my workflow using this tutorial: https://www.youtube.com/watch?v=70H03cv57-o
I hope this helps!
Nice can you do Chobbie and Ryukko?