import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("nubby/kl-f8-anime2-blessed", dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]These VAEs are modified versions of the popular kl-f8-anime2 vae.
These VAEs should not produce a NaN in VAE error even when used on half precision.
They have been modified using the VAE-BlessUp script to produce lower contrast images than the original version.
The recommended version is WD1-4-kl-f8-anime2-bless09.safetensors
Other versions are provided to accomodate different models and different tastes. If you typically prefer the very low contrast of the NAI/Anything vae then you may want to try the 08 version.
WD1-4-kl-f8-anime2-bless08.safetensors = 80 contrast of original/20% decrease
WD1-4-kl-f8-anime2-bless09.safetensors = 90% contrast of original/10% decrease
WD1-4-kl-f8-anime2-bless1-05.safetensors = 105% contrast of original/5% increase
WD1-4-kl-f8-anime2-bless1-1.safetensors = 110% contrast of original/10% increase
- Downloads last month
- 321