Modified Anima for LoRA training.
It reduces unintended style changes.
This is not designed for Anima Yume. Please use Any Anima Yume instead.
Note: This model is not suitable for image generation. The model tends to generate ugly images.
LoRA学習用のAnimaです。
意図しない画風の変化を軽減し、精度も改善します。
注意: このモデルは画像生成に適していません。モデルは醜い画像を生成する傾向があります。
Dataset information:
Dataset size: 5120
Gelbooruから収集した5120枚の画像で学習しました。
なお、以下のタグがあるか条件を満たす画像は除外しました。
filetype:gif, score:<0, mpixels:<1048576, tagcount:<16, \*_artifacts, adversarial_noise, greyscale, monochrome, digimon, photophop_(meidum), ai-generated, duplicate, bad_\*, off-topic, cropped, resized, reversed, rotated, third-party_edit, screenshot, tagme, real_life, watermark, 3d, koikatsu_(medium), mikumikudance, twitter_username
Description
Target version: Preview 3
Recommended resolution: 1024px(1Mpixels)
FAQ
Comments (9)
いつも学習用たすかる～
My LoRa training test results:
(1) character : Preview2 = Preview2.1 > Preview3 > Preview1 (Preview 3 requires a longer training time for characters with very unique traits. More emphasis is placed on aesthetics than on faithfully recreating the character.)
(2) background : Preview3 > Preview2 = Preview2.1 > Preview1 (Preview 3 reproduces the background lighting and shadows very well. But sometimes there is a black background. The official training methods may be aimed at reducing pollution.)
P.S. Preview2.1 has a slightly higher error rate than Preview2 in both training and generation, but there are occasional pleasant surprises.
can u share the config of the traing file
@Aura_farmmer https://xypher7.github.io/lora-metadata-viewer/
@ghostpaint @ghostpaint how i can run this in google cloab
@Aura_farmmer I have never used Google Colab.
@ghostpaint can we talk about it in discord or somthing likethat?
@Aura_farmmer I'm not good at English, so if it's a live conversation, it would be a waste of your time.