Immich vs PhotoPrism for a Family Photo Library: Face Search, Phone Backup and the RAM Bill

When people leave Google Photos or iCloud, they usually picture the end result: every photo the family has ever taken, on a box in the house, searchable and private. What they do not picture is the first month. The phone that stops backing up when it goes to sleep. The machine-learning job that runs for three days. The disk that turns out to be half the size the library needs.

I moved our family library, around fifteen years of phones and cameras, off Google Photos and onto a home server. I ran Immich and PhotoPrism side by side on the same files for a while before committing. Both are good, both are open source, and both will happily index a large library. But they are built around different ideas of what a photo library is, and that matters far more for a family than for one person with a camera.

For a family that mainly takes photos on phones, choose Immich. It is the closest self-hosted equivalent to Google Photos: automatic phone backup, face recognition, smart search, shared albums and a separate account for each person.

For an existing archive of folders on a disk, especially from cameras, choose PhotoPrism. It indexes the files where they already are, without moving or renaming anything, and gives you a powerful way to browse and search them.

Two different ideas of a photo library

Immich is built like Google Photos. Phones upload their pictures to it; it stores them, generates thumbnails and video previews, recognises faces and objects, and presents everything as one timeline. Each family member has their own account and library, and you can share albums or give a partner access to your whole library. It runs as several containers: the server, a machine-learning service, PostgreSQL and Redis.

PhotoPrism is built around folders. You point it at a directory of originals, which can be your existing archive organised the way you like, and it indexes what is there. It reads metadata, recognises faces, classifies images, reverse-geocodes locations, and builds a searchable library on top of the files without changing them. It runs as one application plus a database, usually MariaDB.

That difference, phone-first versus folder-first, explains almost every other difference between them.

Phone backup is what decides it for families

If your family’s photos come from phones, phone backup is the whole job. A library that only has the pictures someone remembered to copy is not a backup.

Immich’s mobile apps do this properly. Install the app, sign in to your server, choose which albums to back up, and new photos upload automatically. There are settings for Wi-Fi only, charging only, and background backup. Background backup on phones is never perfect, especially on iOS, which limits what apps can do while closed. Opening the app now and then to let it catch up is part of life with any self-hosted photo service. But it works well enough that each family member can set it up once and mostly forget about it.

PhotoPrism has no official backup app. Its mobile experience is a web app for browsing. To get photos from phones into it, you use a third-party sync app that uploads over WebDAV, which PhotoPrism supports, or you sync phone folders with something like Syncthing into the originals directory. It works, and some people like the flexibility. But every family member needs that setup, and when it breaks, it breaks silently.

For a family, that alone usually settles it in Immich’s favour.

Faces, search and the machine-learning bill

Both recognise faces and let you name people, which is the feature families use most: “show me every photo of Grandma”. Both also classify images so you can search for things like “beach” or “dog”.

Immich’s search is the more impressive of the two. It uses a CLIP model, which understands natural-language descriptions, so you can search “kids in the snow at night” and get sensible results. It is the closest thing to Google Photos search you can run at home.

That intelligence costs resources. On first import, Immich runs face detection, face recognition and smart-search indexing on every photo, and generates thumbnails and video transcodes. On a large library and modest hardware, that initial pass can take days, and the machine-learning service needs plenty of RAM while it runs. After the initial import, new photos are processed as they arrive and the load is light. But plan for that first week: give the server enough memory, put the database on fast storage, and do not judge Immich by how it feels while the queue is still running.

PhotoPrism’s indexing is also demanding on first run, and its face recognition and classification take time on a large archive. In my experience it is somewhat easier to run on modest hardware, partly because it has fewer services. The trade-off is search that is good, but less natural than Immich’s.

Your existing archive

Most families do not start from zero. There is a folder of camera imports, a backup of an old phone, a Google Takeout archive, and a drive of photos someone organised carefully by year and event.

PhotoPrism shines here. It indexes that folder structure as it is, keeps your folder names visible, and never moves or renames originals. If you have spent years organising folders, PhotoPrism respects that work. It also handles RAW camera files and sidecar files well, which matters to photographers.

Immich can do it too, through external libraries: you point Immich at an existing folder, and it indexes the files read-only, without copying them. That works well for an archive you want to see in the same timeline as new phone photos. But Immich’s heart is its own managed library, and its experience is built around the timeline, not your folders.

Google Takeout deserves a warning, whichever you choose. Takeout splits dates and descriptions into separate files next to each photo, and importing it naively can scramble the timeline. There are community tools that merge that metadata back into the photos before import. Use one. The details are part of what breaks when you leave Google Photos.

Hardware and the RAM bill

Neither is happy on a tiny machine. For a family library of tens of thousands of photos and some video:

• RAM: give Immich at least 6–8 GB if you want machine learning to run comfortably alongside everything else. PhotoPrism is happier with 4 GB or more during indexing.
• CPU: a modern low-power processor, such as the N100-class chips in many mini PCs, handles both. Hardware video transcoding on Intel’s integrated graphics helps Immich with phone videos.
• Storage: put the database and thumbnails on an SSD. Originals can live on large hard drives. Thumbnails and previews add a noticeable percentage on top of your library size.

A used office mini PC or a modern N100 box is a good match for either. If you are choosing hardware, the trade-offs between the best mini PCs for a home server apply directly here.

Upgrades, stability and backups

Immich grew very fast, and in its early years it warned users clearly that updates could include breaking changes. It has matured since, but it still moves quickly. Read the release notes before upgrading, and do not let it fall many versions behind.

PhotoPrism moves more slowly and has been stable for longer. Its free and paid editions differ: some features, including multi-user support, have been tied to paid memberships. Check what the free edition includes before you plan a family setup around it.

Backups are your job with either one. A photo server is not a backup; it is one more copy on one more disk. Back up the originals and the database, keep a copy off-site, and test a restore. For a family archive, losing the database means losing faces, albums and sharing, even if the photos survive.

Phones or folders: picking for your family

Immich is the one when:

• Most of your family’s photos come from phones.
• You want automatic backup that each person sets up once.
• You want separate accounts, partner sharing and shared albums.
• You want the closest self-hosted experience to Google Photos, including natural-language search.
• Your server has enough RAM for machine learning to run comfortably.

PhotoPrism is the one when:

• You have a large existing archive organised in folders and want it left untouched.
• Many of your photos come from cameras, including RAW files.
• You are mainly one person browsing and curating, not a family uploading from phones.
• You prefer a slower-moving project with fewer services to run.

Some households run both: Immich for the family’s phones, and PhotoPrism over the curated camera archive. That is more to maintain, but it plays to each one’s strength. If you can only run one and your family lives on their phones, Immich is the one that will actually end up holding everyone’s photos.