[enhancement]: pull all files needed to generate upon model installation (setting)

[xiaden]

Is there an existing issue for this?

• I have searched the existing issues

Contact Details

applesmycat - aol.com, xiaden@discord

What should this feature add?

Not going to be able to get fancy with the write up here, as my internet is actually terrible right now, but I want to get this documented.

For the cases of use in an sneakernet / airgapped / secure network, Invoke should have the option to download all related files to generate from upon installing an image. this could either be in model manager, or in the actual settings file. something labeled "don't delay dependencies" or "make available offline". I know this is more of a problem with hugging face and the current workflow, but this is causing a world of pain for me currently. Every single other part of InvokeAI is perfectly content to be offline, so I thought it would be fine, and it ran testing fine when disconnected.

Alternatives

No response

Additional Content

No response

[psychedelicious]

Unfortunately the download functionality is all wrapped up in diffusers. You can provide a local path to the required files, but it's not obvious which files it needs ahead of time. We had pending PR that bundled all of them up with the invoke python package but that was rather unpalatable. Maybe we should revisit that solution.

If I recall correctly, the issue is only a problem for safetensors formatted models. If you convert the models to diffusers format, it should grab the required files during conversion and then not need them afterwards.