Verified: a month ago
SafeTensor
Type | |
Stats | 663 4,675 |
Reviews | (103) |
Published | Aug 12, 2024 |
Base Model | |
Training | Epochs: 38 |
Usage Tips | Clip Skip: 1 Strength: 0.9 |
Hash | AutoV2 54546023E3 |
Lora trained in re-create Kisaragi Gunma's art style.
Version 2:
I re-trained 6 times this lora trying to reduce the hands problems and trying to find a way to train my loras faster (and cheaper) than now. But wasn't possible xd, reduce the steps let the loras so undercooked and if I increase the LR, gets a lot of problems in proportions and deformations... Well... I pruned a lot of noisy and low quality images from the dataset and now this version has much less problems whit deformations in general.
Weight: .8 - 1
Version 1:
This one has a hard NSFW bias in a free for all (?
Weight: .8 - 1