Moebius -- general
lightweight image inpainting: an image and a mask in, the masked thing gone

Copyright 2026 Huazhong University of Science and Technology (HUST Vision Lab)
and vivo AI Lab.

Source:  https://huggingface.co/hustvl/Moebius  (pretrained/diffusion_pytorch_model.bin)
Project: https://github.com/hustvl/Moebius
Paper:   Moebius: 0.2B Lightweight Image Inpainting Framework with 10B-Level
         Performance (ECCV 2026), arXiv:2606.19195

Licence: the authors state terms in two places and the two differ. Both are recorded
         here and neither is preferred:
           - github.com/hustvl/Moebius README: "Both the code and the pretrained
             model weights of Moebius are released under the Apache License 2.0 ...
             Commercial use of the weights and the images produced with them is
             permitted."
           - huggingface.co/hustvl/Moebius model card metadata: license: mit
         Both are permissive and both permit commercial use.

This is upstream's byte stream, verified against the sha256 the Hub records and placed
unchanged. No tensor is altered, renamed, cast or dropped.
           pretrained/diffusion_pytorch_model.bin  sha256 6c9ee08ce467af48767b74416646d82f8a0c1dcfb558c478e635cef0ca8b2ad2

This checkpoint is the denoiser only. It cannot run without the autoencoder published
beside it as torch-fp32-vae, which is a separate work under separate terms -- see the
NOTICE that travels with it.