[Tutorial] Immich remote video transcoding

[erkexzcx]

Description & Prerequisites

Decided to write this in order to help fellow users to transcode their library on remote server (e.g. gaming PC), where their immich server isn't powerful enough, or doesn't have required hardware acceleration.

In this example, immich server is the server that hosts your Immich instance on (basically docker-compose.yml contents). remote server is more powerful hardware, e.g. your gaming PC.

Prerequisites:

• Remote server has to be able to access your immich server directly.
• Remote server has to be able to access your ${UPLOAD_LOCATION} mountpoint/files directly. It means that users who don't store files in remote storage (e.g. via NFS) will have a hard time to set it up, but it's not a problem to me.

So before you begin - understand this if you are using local storage, you might need to mount it via SMB or NFS to your remote server. If you cannot do this - this guide tutorial will not work for you.

Procedure

Perform these steps on immich server:

1. Expose redis (ports: [6379:6379]) and database (ports: [5432:5432]) ports in your docker-compose.yml, so these 2 are accessible from the outside (so your remote server can access it).
2. Edit .env and add this variable: IMMICH_WORKERS_INCLUDE: 'api'
3. Restart your docker-compose.yml stack: docker compose down and docker compose up -d.

Now perform these steps on remote server:

1. Download docker-compose.yml, example.env (rename to .env) and hwaccel.transcoding.yml files. From docker-compose remove everything but immich-server and update how you mount your ${UPLOAD_LOCATION}, so it points to the same physical location as your immich server.
2. Edit .env and update DB password and DB name to match what's on your immich server. Additionally, add these variables: DB_HOSTNAME=<ip_of_immich_server> and REDIS_HOSTNAME=<ip_of_immich_server> and IMMICH_WORKERS_INCLUDE: 'microservices'.
3. Bring it up: docker compose up -d and see if no errors: docker logs -f immich_server (all should be green).

Now go back to Immich WebUI and configure hardware transcoding settings. My remote server has RX 7900 GRE, it supports AV1 encoding & decoding and hardware acceleration is powered by vaapi method, so I enabled it. Now go to jobs and press to transcode all your videos. Once it starts, run docker logs -f immich_server on your remote server to see if any errors (e.g. if your hardware acceleration is working). If no errors in red - congratulations, everything is working. :)

Depending on your use-case, you might want to revert the changes. Security & firewall aspects are out of scope for this tutorial.

[legalimpurity]

Awesome!

[lovelacecreativeco]

Thanks for this!

[Junod972]

thanks a looooooot. I will use these like right now. Because, I need to set up at least 3 workers on 3 machines. And, I am struggling like hell to do even the basic.
So, again, thanks a lot.

[biggt]

I love this but what I haven't seen is the ability to have the main server running both API and microservices AND THEN another machine running the microservices for all ML and transcoding needs.

Is it possible to do something like this:

IMMICH_WORKERS_INCLUDE: 'api,microservices'

Official documentation nor unofficial that I have found mentions the specifics of having multiple containers for microservices and ML

[erkexzcx]

Lol. Check out latest release 😅

AND THEN another machine running the microservices for all ML and transcoding needs.

It's getting there. AFAIK whole immich project is trying to become stateless, so you can even host in Kubernetes with various components and scale as you wish. 🙌 It's getting there, just wait.

[bo0tzz]

It's already there. All of this works just fine, you can run whatever components you want wherever you want.

[traktuner]

It's already there. All of this works just fine, you can run whatever components you want wherever you want.

@bo0tzz
Thats awesome!
Will this also get added to the docs like "remote machine learning"? I would really need this just for "remote transcoding"

[flipflower]

Would be great to combine these two, remote machine learning is working perfectly on my pc, would be great just to add an url to video transcoding (or any other type of heavy cpu task)

[bo0tzz]

@traktuner I think that's already covered in https://immich.app/docs/guides/scaling-immich/

[flipflower]

if i have machine learning already set up on my pc & working, how to point to it for transcoding videos?

[erkexzcx]

What?

[flipflower]

i have to recreate the machine learning docker on my pc?

[erkexzcx]

Yes

[DerHexer]

As I was struggling with the details on SMB mountings, I would like to add my results to this post in order to avoid this hassle for others. I'm using the external library function to bring my pictures into immesh (hence the two volumes):

in the .env file:
UPLOAD_LOCATION=immichvolume1

---

in the docker_compose:

[...]
services:
immich-server:
volumes:
- ${UPLOAD_LOCATION}:/usr/src/app/upload
- immichvolume2:/usr/src/app/external:ro
[....]
volumes:
immichvolume1:
driver_opts:
type: cifs
o: username=SMBUSERNAME,password=SMBPASSWORD,domain=SMBNAME
device: "//SMBIP/docker/immich-compose/library"
immichvolume2:
driver_opts:
type: cifs
o: username=SMBUSERNAME,password=SMBPASSWORD,domain=SMBNAME
device: "//SMBIP/photofolder/"
model-cache: