This is a test hyperrnetwork to see what WD 1.4 (and models based on SD 2.1) is capable of.
This hypernetwork is heavily inspired of Black Souls and was trained on some Sushi Yuusha Toro's artworks (Black Souls's author/artist/dev) and somes Black Souls 2 CGs.
It was trained on the first epoch of Waifu Diffusion 1.4 (Anime) [8cbce067] at 512x512 with a learning rate of 5e-5:149,5e-6:10000,5e-7:20000.
In some cases, adding "sketch" or/and "jaggy lines" to the prompt can improve result.
I recommand to not use WD 1.4's quality modifiers (best quality, normal quality,etc...) with that hypernetwork especially as a negative prompt (worst quality,lowres,etc...)
描述:
-20k steps
-"Shuffle tags by ',' when creating prompts."
-Latent sampling method: deterministic
-Modules: 768, 1024, 320, 640, 1280
-Activation function: linear
-149 images
训练词语:
名称: blackSoulsHypernetwork_wd14AnimeEpoch1.pt
大小 (KB): 118507
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: **Detected Pickle imports (3)**
```
collections.OrderedDict
torch._utils._rebuild_tensor_v2
torch.FloatStorage
```
病毒扫描结果: Success
