Immich vs Google Photos: The First Automatic Photo Backup We Were Willing to Turn On (2026)

The short version

This is not a switching story. Our Google Photos library is empty, and it has been empty since roughly the week we first read what automatic upload does. Every photo on the phone, every screenshot, every meme someone sent, every progress picture, every image you would rather nobody else held a copy of, all of it quietly copied to a company whose rules change and whose models learn from what they hold. We turned it off early and never turned it back on.

The cost of that decision was that for years the phone had no automatic backup at all. A dropped phone meant lost photos. That is the trade most privacy-minded people quietly make and do not talk about.

Immich changed it. It is an open-source photo server that runs on our own machine, does the phone upload and the face recognition and the search, and never sends anything anywhere. The setup is more involved than the marketing suggests, the phone side is easier than we expected, and remote access needs a bit of thought. If you have a home server already, it is the answer. If you do not, read the last section before you buy one for this.

Why Google Photos was never on

Google Photos is genuinely good software. The search is excellent, the sharing is effortless, and the free tier is enough for most people to never think about it. None of that was the problem.

The problem is what "automatic backup" means when you stop and think about it. The phone camera roll is not a curated album. It is every image that has passed through the device: photos of documents, photos of whiteboards, screenshots of conversations, memes forwarded in group chats, progress photos taken for nobody but yourself, and pictures you took at two in the morning that you would not want in any album anywhere. Automatic upload sends all of it, by default, to a server you do not control, under terms that have changed more than once.

Add what has happened to the terms since. Face grouping, object recognition, the "memories" resurfacing old photos, and the broader question of what a company that trains models does with a few billion tagged images. Google has policies on all of this. Policies get revised. We covered the pattern in the cloud storage article: the storage was never the problem, the terms were.

So we did what a surprising number of people do and never admit to: turned off sync, kept photos on the phone, and accepted the risk. For years, the backup strategy was "do not drop the phone." That is not a strategy. It is just the price of not handing everything over.

What Immich actually is

Immich is a self-hosted photo and video server. It runs as a set of Docker containers on your own hardware, gives you a phone app that uploads the camera roll automatically, and provides a web interface with a timeline, albums, face recognition, and search by content. It is free and open source. There is an optional paid licence that funds the project and unlocks nothing you need.

The comparison people make is "Google Photos, but on your server," and it is fair. The interface is deliberately familiar. What is different is where the machine learning runs: face detection, face grouping, and the search that lets you type "dog on a beach" all happen on your own box. Nothing leaves. That is the whole point, and it is what makes it the first automatic upload we were prepared to switch on.

We run it on an Unraid server at home, alongside the file storage and the other self-hosted services covered elsewhere on this site. It is one container group among several, and it has been the one that most changed how we use the phone.

Setup: harder than hoped, easier than feared

The honest split. The server side took more effort than the "one-click install" framing suggests, and the machine-learning parts in particular need attention: model downloads, deciding whether to give it a GPU, waiting for the first pass over the library to finish, and understanding which jobs run when. It is not difficult in the sense of requiring expertise, but it is not an afternoon either, and it is not the kind of thing that works perfectly on the first try. Budget an evening for the base install and another for getting the smart features settled.

Unlike our Nextcloud experience, nothing went badly wrong. No permissions disasters, no locked-out folders. Immich wants its own storage location and manages it, which is simpler than pointing a service at files that other things also use. If you give it a clean directory and leave that directory alone, it behaves.

The phone side is the surprise. Install the app, sign in to your server, pick the albums to back up, and it just goes. On the home network it uploads in the background without being asked, exactly the way you would want it to. That part was easier than we expected and it has not missed a photo since.

The one real complication: away from home

On the home network, the phone finds the server and uploads. Away from it, the phone cannot reach a machine sitting behind your router, and nothing uploads until you get back. For a lot of people that is fine: photos wait a few hours and sync when you walk in the door.

If you want uploads and browsing from anywhere, you need a way in. We use Tailscale, which puts the phone and the server on a private network regardless of where either one is, with no ports opened to the internet. It works well and it is the setup we would recommend, but it is another thing to install and understand, and it is the point at which "self-hosted photo backup" stops being a single app and becomes a small system.

Do not expose Immich directly to the internet to avoid this. A photo library is exactly the thing you least want reachable by anyone who finds the port. Tailscale, or an equivalent, is the answer.

Face recognition and search: it works, and you can guide it

The face recognition is not bad. It finds faces, groups them, and gets most of them right. Where it gets one wrong or splits one person across two groups, you merge them and name them, and it learns. "You can guide it" is the honest description: it does the bulk of the work and you do the last few percent, and after that it is reliable. That is a fair trade for it running entirely on your own hardware.

Search by content works the same way. Typing a description of what is in the photo finds it more often than not, and the misses are the kind you forgive because you know nothing left the house to make it possible.

Compared to Google, it is a step down in polish and a step up in trust, and for us that is the correct direction. If you are the person who never turned Google Photos on, the polish was never available to you anyway.

What Google Photos still does better

Sharing an album. Sending a single photo from Immich is just sending an image, the same as from any gallery. Sharing a whole album with family who are not on your server is where it gets less convenient: you can create a shared link, but the recipient experience is not the frictionless thing Google built, and it depends on your server being reachable from wherever they are. For a family that lives inside Google Photos, that is a real difference.

Beyond that, honestly, nothing we miss. But we never used Google Photos, so we are not the right people to list what a heavy user would lose. If your family albums, your shared libraries and your "on this day" habits all live in Google, moving is a bigger change than this article describes, and you should weigh that more heavily than we do.

What we run

Immich on an Unraid server at home. The phone backs up automatically on the home network. Tailscale for the occasions we want it from outside. Face recognition and search running locally, guided a little at the start and reliable since. Google Photos empty, by choice, for years.

The server was already there for file storage and the rest of the self-hosted stack, so the cost of Immich was the time to set it up and the disk it uses. If you already own the box, the same is true for you.

When to skip this

If you do not already have a server, do not buy one for photos alone. A home server is a thing you own and occasionally have to look after, and this site says that in every self-hosting article because it is true. Immich is the right answer for someone who already runs a box and wants it to do one more genuinely useful job.

If you are happy inside Google Photos, and you have read what automatic upload does and decided you are fine with it, stay. That is a legitimate position and most people hold it. The sharing is better and the effort is zero.

If you want the privacy without the server, the hosted option is end-to-end encrypted storage with photo backup, and Proton Drive is the one we would look at first. We have not run it, so treat that as research rather than lived. It will not do local face recognition, because encrypted storage cannot read your photos, which is rather the point. You trade the clever features for the certainty that nobody else is looking.

And if you are where we were, sync off and no backup at all, do something. A phone with no backup is one drop away from losing years. Immich is what we did. Any of the three above beats nothing.