Immich is an amazing piece of software, but because it holds such personal data I have only ever felt comfortable accessing it via VPN or mTLS. This meant that I could never share any photos, which had been really bugging me.
So I built a self-hosted app, Immich Public Proxy, which allows you to share individual files or full galleries to the public without ever exposing your Immich instance. This uses Immich’s existing sharing functionality, so other than the initial configuration everything else is handled within Immich.
Why not just expose Immich publicly with Traefik / Caddy / etc?
To share from Immich, you need to allow public access to your /api/ path, which opens you up to potential vulnerabilities. It’s up to you whether you are comfortable with that in your threat model.
This proxy provides a barrier of security between the public and Immich. It doesn’t forward traffic to Immich, it validates incoming requests and responds only to valid requests without needing privileged access to Immich.
Demo
You can see a live demo here, which is serving a gallery straight out of my own Immich instance.
Features
- Supports sharing photos and videos.
- Supports password-protected shares.
- Creating and managing shares happens through Immich as normal, so there’s no change to your workflow.
Install
Setup takes about 30 seconds:
-
Take a copy of the docker-compose.yml file and change the address for your Immich instance.
-
Start the container:
docker-compose up -d -
Set the “External domain” in your Immich Server Settings to be whatever domain you use to publicly serve Immich Public Proxy. Now whenever you share an image or gallery through Immich, it will automatically create the correct public path for you.
For more detail on the steps, see the docs on Github.
A simpler way to protect a private service with a reverse proxy is to only forward HTTP GET requests and only for specific paths.
It’s extremely difficult to attack a service with only GET requests.
The security of which URLS are accessible without authentication would be up to immich.
This is exactly the risk I’m wanting to mitigate. Immich is under heavy active development, and I want to abstract away from needing to worry whether the no-auth API URLs are safe to expose publicly.
At the end of the day I feel safer knowing that there is zero public access to any part of my Immich instance, which for me is what really matters.
Immich has a whole set of end-to-end automated tests to ensure they don’t accidentally make public any URLs they went to be private:
https://github.com/immich-app/immich/tree/main/e2e/src/api/specs
As a popular open source project, that would be e glaring security hole.
Using this proxy puts the trust in a far less popular project with fewer eyeballs on it, and introduces new risks that the author’s Github account is hacked or there’s vulnerability in he supply chain of this docker container.
It’s also not true that you “never need to touch it again” . It’s based on Node whose security update expire every two years. New image should be built at least every two years to keep to update with the latest Node security updates, which have often been in their HTTP/HTTPS protocol implementations, so they affect a range of Node apps directly exposed to the internet.
All good points, and Apple has some of the most skilled engineers in the world and The Fappening still happened.
It’s not possible for me to audit everything that’s happening security-wise in Immich, but I can fully understand what’s happening in this small codebase to my own satisfaction. At the end of the day I feel safer knowing that there is no public access to any part of my Immich instance.
I meant that you don’t need to use it to share photos, not that you never need to update your docker containers! 😱 Thanks, I have clarified that.
I don’t know the Immich API, but I’ve seen several REST APIs that used the usual pattern of
GET /api/v1/user/<id> - read user POST /api/v1/user/ - create user ...but also allowed
GET /api/v1/user/<id> - read user GET /api/v1/user/?action=create - create user ...Yup, also some APIs use GET for everything. It’s a pain. And it means that filtering by verb only helps if you’re intimately familiar with the API. And even then, only if you keep up with changes as they happen. So really, only if you’re developing the API yourself.
(another pet peeve of mine is “rest” APIs that use 200 response codes for everything)
Ahhhhh whyyyyy, you’ve got all of these standard response codes made for you, why would you blatantly ignore them like that?!
The only one I think is reasonable is GraphQL. But that isn’t rest, and HTTP is just one of the transport layers it supports.
For anything claiming to be RESTful, it’s a crime.
Yes, there are broken uses of the HTTP protocol verbs where filtering to GET won’t work.
Interestingly the recent D-Link NAS security flaw uses GET.
Good example. It’s true that an even a GET request not designed to mutate data might still fail to validate input, allowing a SQL injection attack or other attack that escalates to the privileges that the running app has.