Weird. I'm interested to know your setup. Did you use MariaDB or the built in SQLite?

I have photoprism running on docker, which is running as a VM on a proxmox server that is a 13 year old Dell Optiplex 9020. Only 6Gb ram dedicated to the VM (2GB initial, ballon to 6GB), and photoprism uses 2 workers. I have a large 50k set of photo's from starting from when my daughter was born till now (17). I have never sorted through these photo's other than to dump them from my phone every 6 months (camera before smartphones become common). Initial import took less than 45 minutes and that was before tweaking settings. Full manual index with all options, still takes less than 45. Initial face recognition was about 33 people, once I tweaked settings, it's recognized 97 people with little input from me other than tagging the name correctly for a few photo's per person.

The only difference in my docker file vs the official are optimized settings, and I'm using a separate VM for the database server as I find multi tendency and backups easier with a VM. (VM only has 2GB ram allocated to it)

I'm extremely happy with it, as I always put off organizing my photo collect due to how massive it is. It's made the task so much easier.

I am however always looking for other tools. I'm interested in give yours a try, just wanted see what could be the massive difference in experiences.

I don't remember specifics, though I can find out; I have it seemed in Kubernetes with hand written manifests. I had to tie Photoprism to a specific node which had had more resources than it's peers; I would have to refer to the manifests to recall which DB I used. All storage was over NFS (meaning the storage speed was limited by my 1Gb network backbone), which probably created some of the resource issues.

Edit: as for my tool, it won't replace Photoprism, its primary goal is to deduplicate your photo library.

Not sure about the memory requirements. My container currenty uses 128MB RES memory. But I also only have a couple thousand images. Not sure how strictly in memory some DB/structures are. I assume importing an existing library takes a lot, as it does the inference/training for face detection and auto-tagging. I've seen that it uses a lot of CPU at least, haven't looked into the memory.

Multi-user is lacking to missing. I'm not 100% sure which state photoprism is in, as I was shopping for a self host service. OTOH I don't care about that for family. I'm not hosting anything for untrusted people, or to default backup everything, but as common library. So I'm fine if everyone sees everything and expect people to play nice.

But I do agree that multi user (and maybe integrated auto-sync for phones) are major missing features. I'm just in the situation that it doesn't really matter to me currently.

That's fair enough. Same situation for me really I don't need the auth functionality.

The way I had Photoprism set up is that mine and my wife's phones would sync images nightly to the server, the photos would be rsyncd from the public facing sync server to an internal one (not in the DMZ) then I'd have the server send me a link to the import page on Photoprism because it lacks the feature to autosync when new files are added; I think having to open the page while on LAN (link was internal only) and click a button was the straw that broke the camels back here for me.

I think I also had something like 10-12 cores dedicated to it, which alongside the RAM use that constantly crept up, and the manual intervention, was just too much for a photo management tool that wasn't really even managing the photos in a way that was useful to us.

Are you planning to make it available in some form?

It'll be open source for sure.

I had considered selling it for a nominal fee too but that comes with additional commitments from me, that I'm not sure I want to take on.

Maybe I'll just add Github Sponsors, KoFi or such and see if there's interest, when it's ready and released.