Immich vs Nextcloud for photos

Say the uncomfortable thing first: Nextcloud Photos is not a competitor to Immich, and treating the comparison as close does readers a disservice.

Nextcloud is a file and groupware platform. Photos is an app that draws a timeline over image files that are already in your account, with albums and basic sharing. It is a viewer. Immich is a photo product: mobile apps built for background upload, face clustering, natural-language search over the library, memories, per-item sharing, duplicate detection, and a data model where a photo is the primary object rather than a file that happens to be a JPEG.

The interesting question is not which is better at photos. It is whether you need a photo product at all, because plenty of households genuinely do not.

What Nextcloud actually gives you for photos#

More than the dismissive version suggests, and it is worth being precise.

The Nextcloud mobile client does auto-upload. New camera roll items land in a server folder in the background, and that is the single feature people assume is missing. If your requirement is "the photos are off the phone and on my disk", Nextcloud satisfies it today with an app your family already has for files.

What you do not get is the rest of the product. The built-in Photos app is slow to browse at scale and has no face grouping or semantic search. The usual answers are community apps: Recognize for face and object recognition, Memories for a much better timeline. They work. They are also app store surface, and that is the cost line nobody puts in the comparison table: Nextcloud ships three majors a year, each supported for twelve months, you cannot skip one, and any third-party app not marked compatible with the incoming major gets disabled during the upgrade. Building your photo experience out of community apps means every upgrade window is also a photo outage risk.

Immich's equivalent risk is different: majors break things deliberately, but they arrive once or twice a year and the release notes tell you exactly what changed.

What the spec table is telling you#

Memory in use is the row to read: 3500 MB against 1800 MB. That is not Immich being wasteful, it is the machine learning container holding the CLIP and face recognition models resident. If you want search and faces, you pay it. If you do not, you do not need Immich.

Operational load at 3 against 4 is the row that surprises people. Nextcloud is the heavier system to keep alive despite the lower memory number, because the release calendar, the PHP version coupling and the app compatibility matrix are ongoing obligations rather than one-time setup. Immich's four containers are more services and less homework. Stack planner will hold both figures against the rest of your box.

When running Nextcloud alone is the right call#

This is the case the internet under-argues, so here it is properly.

Run Nextcloud alone if photos are a minor part of a general problem. You want documents, calendar, contacts, shared links for the family, and the camera roll backed up so nothing is lost when a phone dies. You browse photos occasionally, by folder or by month, on a laptop. Nobody in the house has ever typed a search like "beach 2019" and expected it to work.

Run Nextcloud alone if your hardware says so. Immich's documented floor is 6 GB with 8 GB recommended, its Postgres directory must be on local disk rather than NFS or a mapped SMB share, and the amd64 machine learning image needs an x86-64-v2 CPU since v3.0.0, which strands older Xeons and Atoms. On a 4 GB NAS you can run Immich with IMMICH_MACHINE_LEARNING_ENABLED=false, but that removes search and face grouping, which is most of the reason to run it. If that is your situation, Nextcloud plus Recognize is a better use of the same memory. Choosing home server hardware has the sizing.

Run Nextcloud alone if one login, one backup and one upgrade window is worth more to you than face search. That is a legitimate trade and people who make it are not settling.

When you need Immich#

Run Immich if the photo library is the point. The tells are specific: more than one phone uploading, a library past about 30,000 images, anyone who expects to find a photo by describing it, or a household migrating off Google Photos and comparing the result to what they had. Replace Google Photos walks that migration, and the honest answer at the end of it is Immich, not a Nextcloud app.

The other tell is people. Immich's phone app is something you install on a relative's device once and never discuss again. The Nextcloud client is a file sync app with an upload setting, and the difference shows up the first time somebody has to be talked through re-enabling it.

Running both without duplicating your storage#

The common setup is Nextcloud for files and Immich for photos, and the mistake is letting both store the same images twice.

The cleaner arrangement is one owner per file. Either Immich owns photos in its own UPLOAD_LOCATION/library/ tree and Nextcloud never sees them, or Nextcloud owns the folder and Immich indexes it as an external library, read-only, referencing the files in place. The second option is attractive if your family already uses Nextcloud auto-upload and you do not want to change their phones.

Two things to know before you do it. External library folders are not backed up by Immich's own data surface, so they need their own coverage. And if Nextcloud moves or renames a file, Immich's reference goes stale until the next scan. Do not have both applications writing to the same directory.

Whichever way round, keep the databases separate and back them both up as dumps rather than volume snapshots. For Immich that is pg_dumpall plus the library/ tree, restored database first. For Nextcloud it is occ maintenance:mode --on, then config/, the data directory and a pg_dump or mariadb-dump --single-transaction taken together, or the files and the database will disagree about which uploads exist. Backing up a running database covers why that matters more than it sounds.

What happens on upgrade day#

Nextcloud: majors every four months, twelve month support, no skipping. Version 32 landed 27 September 2025, 33 on 18 February 2026, 34 on 9 June 2026. The documented path is latest point release of your current major, then the next major's latest point release, one at a time. Each step is gated on PHP (32 wants 8.1 to 8.4, 34 wants 8.2 to 8.5) and on every installed app being ready. Expect to run occ db:add-missing-indices afterwards, as the web server user, never as root.

Immich: semantic versioning since v2.0.0, so any 2.x app talks to any 2.x server and minor upgrades are boring. Majors are not. v3.0.0 (2 July 2026) removed pgvecto.rs entirely, so instances predating 1.133.0 must migrate to VectorChord first, and downgrades are unsupported even within a minor because migrations run forward only at startup.

The practical rule for running both: never upgrade them in the same window. You want to know which one broke. An update strategy that does not lose data is the routine.

The verdict, by situation#

What to do next#

Test the claim rather than the argument. Install the Memories and Recognize apps on your existing Nextcloud, let Recognize finish an indexing pass, and try to find three specific photos from five years ago. If you find them quickly, you do not need Immich and you have saved yourself 3.5 GB of RAM and a second backup target.

If it is slow, or the faces are wrong, or you find yourself scrolling by month, install Immich with 8 GB available, DB_DATA_LOCATION on an SSD, and MACHINE_LEARNING_WORKERS left at 1. Point it at a copy of one year of photos first and watch memory during the initial import, because that is when hosts fall over.

If the wider question is really about file sync rather than photos, Nextcloud vs Seafile covers the suite-versus-engine split and Seafile is the faster answer for sync alone. Document scanning has its own page in Paperless-ngx vs Nextcloud. The rest of the category is in Photos and files.