Instructions to use AuraDiffusion/16ch-vae with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use AuraDiffusion/16ch-vae with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("AuraDiffusion/16ch-vae", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
Commit History
-
Correct latent scalig factors.
Verified signatureCurrent scaling factor of AuraVAE (`0.18215`) is very wrong. This PR corrects them with adding mean and std values per channel and setting the average scaling factor to the correct average. AuraVAE's average std is very close to 1 and average mean is very close to 0, so both average scaling factor and average shifting factor can be safely ignored (aka set them to 1 and 0). It is better to use per channel scaling / shifting with AuraVAE because of channel 3 being an outlier compared to the rest of the channels. Here are the statistics computed over Danbooru id 8m to 8.5m resized to around 256x256 while keeping the aspect ratio. ``` ------------------------------ Latents Mean: [0.2539, 0.1431, 0.1484, -0.3048, -0.0985, -0.162, 0.1403, 0.2034, -0.1419, 0.2646, 0.0655, 0.0061, 0.1555, 0.0506, 0.0129, -0.1948] Latents STD: [0.8123, 0.7376, 0.7354, 1.1827, 0.8387, 0.8735, 0.8705, 0.8142, 0.8076, 0.7409, 0.7655, 0.8731, 0.8087, 0.7058, 0.8087, 0.7615] ------------------------------ ----…a627a04
-
Update README.md
Verified signature518cefd -
disable force upcast
Verified signature571b815
-
set shift factor to 0
Verified signaturec433fac -
Create LICENSE.md
Verified signatureaff3eed -
Create README.md
Verified signature3236a17 -
Upload folder using huggingface_hub
Verified signatureb1acf3c -
initial commit
Verified signaturef360ceb