What both of these actually do

Neither of these is just a photo gallery you point at a folder. Both are trying to be a full replacement for Google Photos or Apple Photos: automatic backup from your phone in the background, a searchable library with face and object recognition, albums and sharing, and a web UI plus (for Immich) a real mobile app. That’s a meaningfully bigger job than something like a Samba share full of JPEGs, and it’s why both projects lean on a real database and, increasingly, machine learning models running locally to do recognition and search.

Immich: the fast-moving, mobile-first option

Immich is the newer of the two and it shows in the pace of development, releases land constantly, and the feature list has grown fast: a genuinely good mobile app for iOS and Android with background auto-backup that behaves a lot like the Google Photos app people are trying to replace, face recognition and clustering, object and scene search (“find photos of a dog on a beach” actually works), live photo support, and shared albums with other Immich users or via public links.

The trade-off is what you’d expect from a young, fast-moving project. Immich depends on PostgreSQL with the pgvector extension for its ML-powered search, plus Redis for job queues, so the stack is heavier than a single container, docker-compose is the supported and easiest path, and it pulls in several services (server, machine learning service, database, Redis, and typically a reverse proxy in front of it). Breaking changes between versions have happened, and the project is explicit that you should read release notes before upgrading, not just pull the latest tag on autopilot. If you want the closest thing to a drop-in Google Photos replacement, especially the mobile app experience, Immich is currently the stronger pick, but it wants a little more care as an operator.

PhotoPrism: the calmer, library-first option

PhotoPrism takes a different philosophy from the start: your photo library is the source of truth, and PhotoPrism indexes it rather than importing and reorganizing it. Point it at a folder structure you already control (say, a NAS share organized by year), and by default it leaves your files exactly where they are and builds a searchable index on top, rather than moving everything into its own managed storage layout. For anyone who’s been burned by a tool that silently reorganizes or renames a decade of files, that’s a meaningful reassurance.

It also does face recognition, object/scene tagging via TensorFlow models, and a clean web-based gallery with albums, search, and sharing. Where it’s behind Immich is mobile: there’s no dedicated first-party mobile app with background auto-backup the way Immich has one, you’re generally relying on a third-party sync tool (or manually copying files onto the share PhotoPrism watches) to get phone photos into the library in the first place. Some of PhotoPrism’s more advanced features, like certain search and organization tools, sit behind a paid “Plus” membership rather than being fully free, which is a legitimate way for a small team to fund development, but worth knowing going in since Immich’s full feature set is free.

The differences that actually matter for picking one

Mobile backup. If the whole point is “my phone photos land somewhere safe automatically, the way Google Photos did,” Immich’s mobile app is the deciding factor. PhotoPrism can get there, but it needs an extra piece (Syncthing, a scheduled sync script, or manual imports) rather than having it built in.

How much you trust your file organization already. If you already have a clean, backed-up folder structure and just want search and a nice gallery layered on top without anything touching the files themselves, PhotoPrism’s library-first model is the safer, more conservative choice. If you’d rather hand the whole problem, ingestion, organization, and backup tracking, to the tool and not think about it, Immich’s managed approach fits better.

Stack complexity and your appetite for upgrades. Immich’s Postgres/pgvector/Redis stack is more moving parts, and its faster release cadence means more frequent upgrades to actually track. PhotoPrism’s stack is simpler and the pace of breaking changes has historically been slower. If you want to set it up once and mostly leave it alone, that favors PhotoPrism. If you’re fine treating it like any other actively-developed self-hosted app you keep current, Immich’s pace isn’t a real downside.

Cost of the full feature set. Immich’s entire feature set, including ML search, is free and open source. PhotoPrism’s core is free too, but some organization and search features are gated behind a paid membership. Neither is unreasonable, just budget for it if PhotoPrism’s Plus tier features turn out to matter to you.

Where to actually run it

Both are comfortable in a dedicated LXC or VM on a Proxmox homelab, and both benefit from being kept separate from your bulk media storage rather than crammed onto the same volume as your Plex or Jellyfin library, photo libraries and their thumbnail/cache generation do enough small-file I/O that a dedicated data path (or at least a dedicated dataset if you’re on ZFS) keeps things tidier. ML-based face and object recognition runs on CPU by default for both projects and works fine that way for a personal-scale library; a GPU speeds it up but isn’t required unless you’re indexing a genuinely huge library and want recognition to keep up in near real time. Give either one persistent storage that survives a container rebuild, the photo library and the database both need to live outside the container’s own ephemeral filesystem, not just inside it.

Don’t confuse “self-hosted” with “backed up”

This is the trap worth calling out directly: moving your photos off Google Photos and onto a self-hosted tool does not automatically make them safer, it just moves where the single point of failure lives. A drive failure, a bad rm, a ransomware hit, or a Proxmox host that dies without warning can take out a self-hosted photo library just as completely as a cloud account getting locked, the difference is nobody’s paying an SLA to help you recover it. Whichever tool you pick, the actual data needs to live under the same 3-2-1 backup discipline as the rest of your data-hoarding setup, multiple copies, on different media, with at least one copy off-site, not just “it’s on my NAS so it’s fine.” A photo gallery app is not a backup strategy on its own, no matter how good its search is.

Bottom line

Pick Immich if you want the closest thing to a real Google Photos replacement, especially the mobile app with automatic background backup, and you’re comfortable running a slightly heavier stack and keeping up with a fast release cycle. Pick PhotoPrism if you already trust your file organization, want a calmer and simpler stack that mostly runs itself, and don’t mind handling phone backup as a separate step. Either one gets your photo library off a third party’s servers and under your own control, which is the actual goal, the choice between them is about how much you want the tool to manage for you versus how much you’d rather manage yourself.