As someone who loves immich, a VM is overkill but also without vGPUs or external configuration loses you access to some the best features, local AI searching akin to what google and apple photos offer.
Its not perfect but its great to be able to just search for things in a photo and find any matches across dozens of TBs of raws, without having to have some 3rd party cloud AI nonsense do all the work.
The only thing I wish they could get integrated is support for jxl compressed raws, which requires them compile libraw with adobe's sdk.
It seems like you are saying the AI features don't work if you don't have a GPU, if I understood correctly, but I have my install on a server with no GPU and the object search and facial recognition features work fine. Probably slower to generate the embeddings, but I don't have any comparison to make.
> As someone who loves immich, a VM is overkill but also without vGPUs or external configuration loses you access to some the best features, local AI searching akin to what google and apple photos offer.
I installed immich in a VM. And the VM is using GPU passthrough. I don't see how it's overkill: immich is a kitchen sink with hundreds if not thousands of dependencies and hardly a month goes by without yet another massive exploit affecting package managers.
I'm not saying VM escapes exploit aren't a thing but this greatly raises the bar.
When one install a kitchen sink as gigantic as immich, anything that can help contain exploits is most welcome.
So: immich in a VM and if you want a GPU, just do GPU passthrough.
I agree that the search using facial recognition is nice in immich that said.
Its not perfect but its great to be able to just search for things in a photo and find any matches across dozens of TBs of raws, without having to have some 3rd party cloud AI nonsense do all the work.
The only thing I wish they could get integrated is support for jxl compressed raws, which requires them compile libraw with adobe's sdk.