[Issue]: Double download when using models from Huggingface

[Tillerz]

Issue Description

When you download a model using the HF download inside SDNext, files get downloaded to models/Diffusers/. When you select/activate the model after download, the files get downloaded again, this time into models/higgongface/hub/.

That takes twice the time, bandwidth and diskspace.

Possibly the same for other SDNext internal downloads (CivitAI)?

Version Platform Description

Windows 11, WSL-2, Ubuntu-Box

Relevant log output

Backend

Diffusers

Compute

nVidia CUDA

Interface

ModernUI

Branch

Dev

Model

Unspecified

Acknowledgements

• I have read the above and searched for existing issues
• I confirm that this is classified correctly and its not an extension issue

[ryanmeador]

I've observed this as well on Linux. I didn't think much of it because I've messed with the default location for the HF cache, but maybe it's a broader problem.