I don't think there are services like that, since usually this means deploying and destructing an instance, which takes a few minutes (if you just turn off the instance you still get billed).
Probably the best option would be to have a snapshot, which costs way less than the actual instance, and create from it each day or so yo run on the images since it was last destroyed.
This is kind of what I do with my media collection, I process it on my main machine with a GPU, and then just serve it from a low-power one with Jellyfin.
create from it each day or so yo run on the images since it was last destroyed.
Unfortunately, for this usecase, the GPU needs to be accessible in real time; there is a 10 second window when an image is posted for it to be processed [1].
fedi-safety must run on a system with GPU. The reason for this is that lemmy provides just a 10-seconds grace period for each upload before it times out the upload regardless of the results. [1]
You can actually run it in async model without pictrs safety and just have it scan your newly uploaded images directly from storage. It just doesn't prevent upload this way, just deletes them.
Probably the best option would be to have a snapshot
Could you point me towards some documentation so that I can look into exactly what you mean by this? I'm not sure I understand the exact procedure that you are describing.