PixelHacker VAE -- the autoencoder Moebius denoises inside
512x512 pixels <-> a 64x64x4 latent

Copyright Huazhong University of Science and Technology (HUST Vision Lab).

Source:  https://huggingface.co/hustvl/PixelHacker  (vae/diffusion_pytorch_model.bin)
Project: https://github.com/hustvl/PixelHacker
Paper:   PixelHacker: Image Inpainting with Structural and Semantic Consistency,
         arXiv:2504.20438
Licence: MIT, stated on the Hugging Face model card.

Provenance worth stating plainly: the config this ships with is Stable Diffusion XL's
VAE config with sample_size changed from 1024 to 512. Every other value matches,
including scaling_factor 0.13025 to five digits, so this is SDXL's autoencoder,
fine-tuned or not. That matters only for terms, and it does not change them:
stabilityai/sdxl-vae is published as its own MIT repository, separate from SDXL base's
OpenRAIL++. The chain terminates in MIT whichever link is followed.

Upstream stores these tensors in half precision; they are placed here unchanged, and
mozo casts to fp32 at load.
           vae/diffusion_pytorch_model.bin  sha256 a59d7ea697f2942d22002dc3469e8c53db807a6b78f7f5ec03bd4c1f70f98efe