AI based photo search plugin?

[danergo 37]

On recent phones there is a nice ai search tool in gallery: you can search for people, pets, cars, documents etc.

Is something like this planned into emby?

Like for camera uploads for example. Or do we already having this?

[Luke 42077]

For those who gave this a thumbs down, what is the concern?

[RanmaCanada 494]

6 minutes ago, Luke said:

For those who gave this a thumbs down, what is the concern?

Far too many things on the developer's plates. Let this AI crap die as far too many users will attempt to use it on their potato servers that absolutely can not handle the workload, and then they will complain to high heaven that the Emby dev team is doing a horrible job of supporting said plugin. We've seen it time and time again where users who know nothing about their hardware or playback requirements complain and argue with us experts when we inform them that they are the problem not Emby. And instead of understanding they double down and embrace their ignorance and get supported by similar users who are completely clueless. We get angry, we block users, and mods have to intervene to delete threads and warn users.

If you're going to do something like this, make it perfectly clear that you will not support users unless they have the bare minimum of hardware/software requirements. And even then, users will still complain that you're being discriminatory to those who can't afford fancy video cards with tensor cores, and will demand that figure out a method of use their potato hardware "or else". Already seeing this with people requesting realtime AI subtitles and stating that using CPU is too slow...DUH.

[danergo 37]

My emby is running on a server with a dedicated GPU (nvidia quadro). Its a very capable component.

For the rest: AI is resource hungry. So any component shall run only in idle periods.

We can drop an info bar on top  of the gallery that AI hasn't seen all items yet. Hence we can give the user some basic info on what's going on.

It's always a common goal for developers and users to have proper education and polite feedback.

Users might be unaware of what resources are required for running certain features, but this is not entirely their fault. They also might unaware of what resources they actually have.

Therefore these features IMHO shall exists only if certain hw criteria is met.

Similar to transcoding: your hw can do it on certain level. Add a dedicated GPU and you will know how bad was it without.

So back to the topic: AI based categorisation shall run in idle (nights or workhours) period. Thus only new images shall be categorised, so running it periodically seems feasible.

So let me repeat my question: is this on the future timeline of Emby?

 

[Suliamu 36]

9 hours ago, Luke said:

For those who gave this a thumbs down, what is the concern?

There is other projects who do exactly that, and only that. 
Very well developed with hundreds of contributors and popular among users:   
https://github.com/photoprism/photoprism 
I have to say that i am using emby for video, and for video only. I see that there is the drive to make this a universal solution for everything (audio, foto, games, books), but in order to do that successfully you would have to be better than all those other particular projects that already do that. This is a lot of development time that would be going into something that already exists. And from the perspective that there is only a handful people working on emby i have to give a thumb down out of my own interest. 

Edited October 22, 2025 by Suliamu

[Gilgamesh_48 1240]

9 hours ago, Luke said:

For those who gave this a thumbs down, what is the concern?

I do not like or want anything AI in my media management. It is simply because AI is too flawed and allowing AI any access at all is potentially an extreme hazard. There is just too much "A" and not near enough "I" in "AI."

All my interactions with anything AI have been disappointing at best.
Humans are better. Maybe not a lot better but still better.

[RanmaCanada 494]

11 hours ago, danergo said:

My emby is running on a server with a dedicated GPU (nvidia quadro). Its a very capable component.

For the rest: AI is resource hungry. So any component shall run only in idle periods.

We can drop an info bar on top  of the gallery that AI hasn't seen all items yet. Hence we can give the user some basic info on what's going on.

It's always a common goal for developers and users to have proper education and polite feedback.

Users might be unaware of what resources are required for running certain features, but this is not entirely their fault. They also might unaware of what resources they actually have.

Therefore these features IMHO shall exists only if certain hw criteria is met.

Similar to transcoding: your hw can do it on certain level. Add a dedicated GPU and you will know how bad was it without.

So back to the topic: AI based categorisation shall run in idle (nights or workhours) period. Thus only new images shall be categorised, so running it periodically seems feasible.

So let me repeat my question: is this on the future timeline of Emby?

 

Just because you have a Quadro card doesn't mean you have Tensor Cores.

As per Luke's response, you can infer this is obviously not on the timeline at all, and it is not something the majority of Emby users are interested in. Very few people use Emby to look at pictures. Heck we're still waiting for the promised comic book reading capabilities cbr/cdz/pdf and e-reader capabilities that was introduced with 4.9.1.8 to work properly. We've waited almost a decade for this to be implemented, no seriously, a decade. The dev team has far too many projects on the go. Let AI die. It's useless.

[Luke 42077]

I think regardless of how you feel about AI, i think the ability to recognize a dog in a photo and then filter for all photos with a dog but would be neat.

[Gilgamesh_48 1240]

3 minutes ago, Luke said:

I think regardless of how you feel about AI, i think the ability to recognize a dog in a photo and then filter for all photos with a dog but would be neat.

As long as it can be and defaults to being completely turned off then it is OK. It is not OK if it is forced on users or if it requires any action to avoid. 

I have never met an AI that showed any  "I" but they all show a LOT of "A." 
I also think it is truly valueless but some people seem to like it and I just want any AI functionality to NEVER be seen by me and to NEVER slow things down.

[Lessaj 467]

I think this thread is enough evidence that no one actually want this, not a single reaction agreed with OP, and all agreed on the other posts saying not to do it and/or there are other products for this. Development efforts are far better spent on actual core features and bug fixes.

[yocker 1247]

I like the idea, would be a useful and fun addition.
I just don't think the old paper weight of a GFX card that i have in the server would be able to handle it and transcoding at the same time.
It would require a beefy server, mobile phones are fast because the AI stuff usually happens in the cloud and as Emby is a personal computer i personally don't think cloud based AI would be fit well.
If some Google, ChatGPT integration is made then it should warn in large letters that the data is shared and not private!

Other than that i as said like the idea.

[cptlores 40]

"a jack of all trades, master of none"..

So focus on the core video hosting features and make improvements related to that.

Install Immich or PhotoPrism if you want a proper solution focused on handling large photo collections.

Edited November 4, 2025 by cptlores

[Mollyplays 2]

I know this thread is a bit old, but I’m curious if anyone has tried hooking Emby up with external AI photo-tagging tools in the meantime. Some self-hosted options like Immich or PhotoPrism can auto-tag images—maybe there’s a way to sync or import tags back into Emby? Would love to hear if anyone has experimented with workflows like that or if there’s been any movement on a native plugin.

[Mollyplays 2]

I’ve seen folks handle this gap by running their photos through tools that tag faces or objects before Emby scans them. Something like Social Media Reverse Image Search can give you quick face matches, which you can then use to organize folders more cleanly. It’s a bit of a workaround, but until Emby rolls out anything built in, pre-tagging your camera uploads can save you some headaches.

Edited January 16 by Mollyplays

[rbjtech 5284]

Anybody that has used Amazon Photo's facial recognition to find all the photo's of a particular person will tell you it is pretty amazing and a massive time saver when you have thousands of photo's.   It clever enough to find the same person from very early age to adult - dislike AI maybe, but that is prety impressive (and useful) imo.