Verified: 9 months ago
SafeTensor
Type | |
Stats | 340 5,291 1.7k |
Reviews | (60) |
Published | Jul 30, 2024 |
Base Model | |
Training | Steps: 15 Epochs: 15 |
Trigger Words | by araki hirohiko |
Hash | AutoV2 9622F2BDA9 |
The illustrious version LORA was trained on 20 Hirohiko Araki artworks to preserve the style characteristic of his original drawings rather than the anime adaptation. Unfortunately, even with a small training step, it managed to overfit. Because of that, I don’t recommend using a CFG higher than 3 — anything above that already looks too garish for my taste and strays from the original intent.