from shellington@piefed.zip to selfhosted@lemmy.world on 14 Jun 15:06
https://piefed.zip/c/selfhosted@piefed.zip/p/1575920/think-i-am-getting-carried-away-with-self-hosting-stock-image-library
I have nearly every service imaginable running and have now started a new project.
I am creating a searchable stock photo archive for my lan. It has been a very interesting project but think i may have crossed the line into overkill lol.
I had hundreds of stock photo cds from the 90s I have turned them all into ISO’s.
I then spent ages dealing with some strange cdrom layouts but got all the images off.
I then converted them all to JPG.
I have now setup a batch script that dedupes then takes the images in 2k batches, runs them through a ai vision model to add keywords and descriptions; as they have none.
They are then copied to a folder where I have photoprism running as the front end and I only have 4k done so far but they look amazing and the search and descriptions are really accurate and useful.
400k more images to go but at least it should all be automated now.
#selfhosted
threaded - newest
Careful, once it’s automated you won’t be able to work on it anymore!
This feeds into my ultimate project which I think will take me all summer.
I plan to create a lan wide search that has kiwix, tube archivist, ubooquity, paperless, jellyfin, stock images and stash.
Then I would have a unified search point and i think it would make kiwix far more usable by not having to go to the specific zim first. BBut, it is a tricky project as some things are nice and have api’s others don’t
Are you a web designer, or how do you utilize 400k+ of stock images?
No i’m not just kind of thought it would be nice to have, that way if i ever need to make any cards or banners i have loads of stuff in every category, with no ugly AI pics like most search engine image searches show these days.
Well, that’s pretty cool.
You can use Immich, it’s not perfect for this use case but it is searchable by content
Might be worth a try.
The thing that has surprised me the most is how good this AI model is at accurately knowing what is in a picture when the model itself is only 3GB in size.
How do you do the AI tagging? That is something I need right now.
running the images through ollama using this model: gemma3:latest
It is working on cpu only and still gives a decent throughput
Cool, thanks. I’ll look into it.
That’s awesome! But… why?
kind of getting ready for the day when the open web becomes pretty much unusable due to ai and id requirements
And when the slopocalypse and technofascism have blown over, we crawl out of our digital bunkers and repopulate the wastelands of cyberspace with… stock images of Hide the Pain Harold? I love it! 😄 Although I could think of more valuable data to hoard for that event.
Any chance you can upload your ISOs to archive.org?
This is awesome!
Personally I would have used TIFF and either Immich or ResourceSpace (a DAM - meant for this kind of thing, but also maybe more institutional than you want.)
Damn, unlimited storage? In this economy?