Description
Initial.
The v0.0.0 was trained on 100 images in total:
year 2026: 34 imagesyear 2025: 20 imagesyear 2024: 46 images
Because the training dataset includes Yordles, the LoRA tends to generate characters with four fingers.
This can be somewhat mitigated by adding yordle or fewer digits to the negative prompt, or by using NegPip.
dim32/alpha32, CAME, UNet LR 3e-05, cosine 900 steps, 100 images (1024²)