Hi all,

 

i want to use Immich on my Unraid Server but I can't find any "machine learning" docker for Immich. How I can use the "machine learning" part locally on my Intel NUC?

 

Thank you and Regards
Luk

the machine learning sytem is built into immich.

immch maintainers want us to use the docker compose plugin to run on unraid:
https://immich.app/docs/install/unraid/

 

space invader has made CA unraid docker tempaltes.
You will need to run

 

its scattered on the forum but i've assisted other with setting up immich.

I'm currently running it with a nvidia and use the buffal_s modle for face detection.

The redis docker mod is broken and breaks when immich has updates.

Machine learning is possible via the setting once its up and running:

 

If you don't want to self-host the machine learning you will need to use the docker compose version and setup a remote access to something like chatgpt / openai for your instance
https://immich.app/docs/guides/remote-machine-learning/

@bmartino1 is it better to run a redis container versus using docker-mods? 

 

Additionally, would you be able to share your immich container settings because I swear I have them set up appropriately, but nvidia-smi in immich console is blank. 

 

 

 

 

I had issues with immich updates and runign the docker mod. it is better to run a seperate redis container.

extra parmaerter:
 

--runtime=nvidia --hostname immich.home.arpa --device=/dev/dri:/dev/dri

added nvidia stuff:

 

Advance toggle more setting:
this docker runs as nobody:users with a umask of 022

 

Machine Learning and G-Card at bottom:

Redis info:

 

Postgress info:

 

Immich settings / volumes:

 

using repository - docker tag:

ghcr.io/imagegenius/immich

 

Thank you so much, I'll check it out and fingers crossed my GPU can start carrying its load

if the g card is being used by other containers, you may need to edit immich machine learning settings and move to a smaller ai. I found success with buffalo_s for facial recognition.

 

check the log for dns / ai grab and memory error:

edit immich settings:

profile > administration:

 

 

settings at the left:

 

Machine learning:

Facial Recognition:

...

Review:

https://immich.app/docs/features/facial-recognition/

 

more info on how and what models are used:

https://pypi.org/project/insightface/0.7/

 

 

Edited April 17, 20251 yr by bmartino1
PI delete

Hey @bmartino1,

I've followed your guide here exactly but my CPU seems to still be the one running the ML models here while my GPU sits idle.

Just to confirm, for your docker image repository and tag, which version are you using?

ghcr.io/imagegenius/immich

or 

ghcr.io/imagegenius/immich:cuda

 

Edited March 9, 20251 yr by mykal.codes
clarification

ghcr.io/imagegenius/immich

 

which is the latest tag. It uses cpu and the nvidia card the machine learning is more setting within immich and another docker...

https://immich.app/docs/guides/remote-machine-learning/

 

you would want to run cuda version to use nvidia exclusively.

Hi, sorry to resurrect this, but aside from pointing the container to the host's /dev/dri , what arguments would be needed on the Intel side?

Must one specifically use the 'openvino' tag to retrieve an Intel-specific image from Imagegenius's repo, or does the 'latest' tag include all versions (cuda, openvino, noml)?

My preferred hardware device appears to be set correctly:

--runtime=nvidia --hostname immich.home.arpa --device=/dev/dri:/dev/dri

EDIT: I think I've found the solution for Intel and openvino:

These are the parameters: --device=/dev/dri --device-cgroup-rule='c 189:* rmw' -v /dev/bus/usb:/dev/bus/usb

And, the repo should have the latest openvino tag: ghcr.io/imagegenius/immich:openvino

Edited January 2Jan 2 by floepie05
To update

Could you, please help me? Hardware transcoding works - I see a load on my 1050ti, but all machine learning tasks a still processed on the CPU. What am I doing wrong?

Edited February 14Feb 14 by DarkP

when isntalling CA are you installing the cuda version?

also folow the nvida plugin 1st post data:

https://forums.unraid.net/topic/98978-plugin-nvidia-driver/#findComment-913251

Edited February 15Feb 15 by bmartino1

On 2/15/2026 at 6:26 AM, bmartino1 said:

when isntalling CA are you installing the cuda version?

I've installed Default version. Need to install Cuda version?

3 hours ago, DarkP said:

I've installed Default version. Need to install Cuda version?

If using the ca, you need to install the cuda version. That's because the default Docker does not have the drivers.

Similar back with any other system you need to install the drivers to use that device. The default does not have the graphics drivers installed. That's why you need to install the appropriate Docker based on your hardware...

On 2/19/2026 at 8:32 PM, bmartino1 said:

If using the ca, you need to install the cuda version. That's because the default Docker does not have the drivers.

Similar back with any other system you need to install the drivers to use that device. The default does not have the graphics drivers installed. That's why you need to install the appropriate Docker based on your hardware...

Many thanks! It works now!