Verified: 2 years ago
SafeTensor
Type | |
Stats | 1,243 |
Reviews | (189) |
Published | Apr 28, 2023 |
Base Model | |
Training | Epochs: 8 |
Trigger Words | kousaka honoka |
Hash | AutoV2 250A196BAF |
Activation tag: kousaka honoka
Recommended ratio: 0.6
Dataset:
92 anime screencaps mostly from S2 and movie (repeat 3)
94 SIF UR both idolized/unidolized (repeat 1)
Training details:
Unet lr: 4e-4
Text lr: 1e-4
Epochs: 8
Base: animefinal-pruned-fp16
This LoRA might occasionally generate distorted eyes from non-close-up views. You can fix it by using img2img impainting with denoising strength 0.3.