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 <a target="_blank" rel="ugc" href="https://twitter.com/toro_yori_ebi">Sushi Yuusha Toro's</a> artworks (Black Souls's author/artist/dev) and somes Black Souls 2 CGs. It was trained on the first epoch of <a target="_blank" rel="ugc" href="https://gist.github.com/harubaru/8581e780a1cf61352a739f2ec2eef09b">Waifu Diffusion 1.4 (Anime) [8cbce067] </a>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...)

Black Souls hypernetwork for WD 1.4

-20k steps -"Shuffle tags by ',' when creating prompts." -Latent sampling method: deterministic-Modules: 768, 1024, 320, 640, 1280-Activation function: linear-149 images

Files with SHA256: cf53ab40f3e9d75e8059217b4c38619e43f311939b08c0d9772ae5c9854af41e