Your Photos, Your Server: Escaping Google with Self-Hosted Immich

Your phone has quietly become the family archivist. Every birthday, holiday, and blurry photograph of a meal you wanted to remember is funnelled, by default, into a cloud service owned by a company whose business is understanding you. Google Photos is genuinely excellent software, but the price of that convenience is handing over an intimate, decades-long record of where you have been, who you know, and what your life looks like. There is now a credible way to keep all of that magic while moving the storage onto a server you control. It is called Immich, and this guide walks through standing it up on a Linux box and pointing your phone at it instead of the cloud.

1 The Quiet Problem With Cloud Photos

The discomfort with cloud photo services is not paranoia, it is arithmetic. A modern photo library is a machine-readable dossier. Embedded in each image is the precise location and time it was taken; layered on top by the service is face recognition, object detection, and the ability to search your entire life by typing “beach 2019”. That is wonderful when you control it and unsettling when a corporation does, because the same indexing that finds your dog also builds a profile that can be analysed, monetised, or handed over under subpoena.

There is also the matter of leverage. Once tens of thousands of irreplaceable images live in a proprietary silo, the friction of leaving is enormous, which is exactly why pricing tiers creep upward and free storage quietly disappears. Owning your photo infrastructure removes that leverage entirely. You decide the storage limit, you decide who has access, and nobody can change the terms on you.

2 What Immich Brings to the Table

Immich is a self-hosted photo and video management platform built deliberately to feel like Google Photos, and it gets remarkably close. It offers automatic background backup from a mobile app, a fast timeline view, albums and shared albums, a world map plotting where your photos were taken, face recognition that groups people automatically, and machine-learning search that lets you find images by their contents rather than their filenames. Type “red car” or “snow” and it surfaces matching photos, all processed on your own hardware.

It is genuinely impressive, and the development pace is fierce. That pace comes with a caveat we will return to: Immich’s own developers ask you to treat it as software still under heavy development and to never rely on it as your only copy of anything. Take that warning seriously and you get an extraordinary amount of capability for nothing but the cost of a server and some disk space.

3 Installing With Docker Compose

Immich is distributed as a set of containers orchestrated with Docker Compose, and the project provides the canonical files. Create a directory and fetch them:

mkdir immich && cd immich
wget https://github.com/immich-app/immich/releases/latest/download/docker-compose.yml
wget -O .env https://github.com/immich-app/immich/releases/latest/download/example.env

Open the .env file and set, at minimum, where your photos will live and a strong database password:

UPLOAD_LOCATION=/srv/immich/library
DB_DATA_LOCATION=/srv/immich/postgres
DB_PASSWORD=use-a-long-random-string-here
TZ=Europe/London

Then bring the whole stack up:

docker compose up -d

This launches several containers working together: the main server, the web interface, a PostgreSQL database with vector extensions, a Redis cache, and a dedicated machine-learning container. Once they settle, visit http://your-server-ip:2283, create the first admin account, and you have a running instance.

4 Pointing Your Phone at It

The real test is whether photos flow in automatically. Install the Immich app from the App Store or Google Play, and on first launch it asks for a server URL and login. Enter your server’s address, exactly as you reach it, and sign in with the account you just created.

Inside the app, open the backup settings and choose which albums to back up. Most people select the camera roll and leave it on continuous background backup. The app will then upload new photos as you take them, the same behaviour you are used to from the cloud, except the destination is your own machine. Let the first sync run on Wi-Fi and on charge, because an existing library of tens of thousands of images takes time and bandwidth to transfer.

A practical note: background backup on mobile operating systems is inherently best-effort, since both Android and iOS aggressively suspend apps to save battery. Open the app periodically to let it catch up, and do not assume a photo is backed up until you have confirmed it appears in the timeline.

5 Planning Your Storage

Photo libraries grow relentlessly, and modern phones produce large files, so storage planning is not optional. A rough way to estimate: a year of casual shooting from a single modern phone often lands somewhere between a handful and a couple of dozen gigabytes, and video inflates that dramatically. Multiply by the number of people backing up and the number of years you intend to keep, then add generous headroom.

Put the UPLOAD_LOCATION on a disk you can grow, ideally a dedicated drive or a RAID array rather than your system disk. Many people run Immich on a NAS or a small server with mirrored drives so a single disk failure does not take the library with it. Whatever you choose, the storage strategy should be decided before you import a decade of memories, not after.

6 The Machine-Learning Container

The feature that makes Immich feel like the real thing is its smart search and face recognition, and these run inside a separate machine-learning container that ships with the stack. When photos are imported, this service generates embeddings that power semantic search and detects and groups faces, all locally. Nothing is sent to a third party.

This is also the most resource-hungry part of the system. On a low-powered machine the initial indexing of a large library can take a long while and will load the processor heavily, so let it grind through in the background. If your server has a compatible GPU, the project documents hardware-accelerated configurations that speed this up considerably. For a typical home setup, a patient CPU-only run during off-hours works perfectly well; you simply wait for the smart features to populate.

7 Backing Up the Backup

Here is the part nobody enjoys but everyone needs. Immich is your primary store now, which means a server failure could wipe the very memories you moved there to protect. A self-hosted photo service is not a backup; it is a single point of failure unless you make it otherwise.

You must protect two things: the photo files at UPLOAD_LOCATION and the PostgreSQL database that holds all the metadata, albums, and face data. The files can be synchronised offsite with a tool like rclone or restic, and the database should be dumped on a schedule. Immich documents the correct procedure, which uses the database container directly:

docker exec -t immich_postgres pg_dumpall --clean --if-exists \
  --username=postgres | gzip > immich-db-$(date +%F).sql.gz

Pair that nightly dump with regular offsite copies of the upload folder, and you have a genuine three-copy strategy. Test a restore at least once, because an untested backup is only a hopeful guess.

8 An Honest Verdict

Immich is the closest thing to a self-hosted Google Photos that has ever existed, and for anyone uneasy about handing their family album to an advertising company, it is a revelation. The mobile backup works, the search is uncannily good, and the whole experience feels polished beyond its years. The one rule you must internalise is that it remains fast-moving software under active development, so you keep independent backups and you do not delete the cloud or the originals until you have proven your own setup over time. Treat it with that respect and Immich gives you something genuinely rare: a lifetime of memories living somewhere that answers only to you.