If you’ve been putting off leaving Google Photos because nothing else felt quite ready, 2026 is probably the year to make the move. The self-hosted options have matured significantly, and for most home server setups, you can match the core features that matter day-to-day — automatic mobile backup, face recognition, browsing by date and place — without sending your family photos to a cloud you don’t control.
This comparison covers the four most practical self-hosted photo managers right now: Immich, PhotoPrism, Nextcloud Memories, and LibrePhotos. I ran all four on the same hardware over several months, tested mobile apps, ingested the same 40,000-photo library, and pushed each one for sharing and multi-user workflows. Here’s what actually matters.
What to Look For
Before diving in, a quick checklist of what separates a usable daily driver from something you’ll abandon in three months:
- Mobile backup: Must be automatic, reliable, and battery-friendly. If it fails silently, it’s useless.
- Face recognition: Accuracy varies enormously. All four use machine learning, but the quality and management UI are very different.
- Resource use: RAM and CPU during indexing will surprise you if you’re running on a low-power box.
- Multi-user support: Sharing an album vs. separate vaults for each family member are very different features.
- Active development: Photo management software that isn’t actively maintained will fall behind mobile OS changes and eventually break backup apps.
- Migration path: Can you get your photos back out cleanly if you switch?
—
Immich
Immich is the closest thing to a direct Google Photos replacement available right now, and it has earned that reputation by shipping fast and listening to users. The interface is a clean, responsive timeline that loads quickly even with tens of thousands of photos. Mobile apps (iOS and Android) are polished and handle background upload reliably — this was the first thing I tested and the first thing many alternatives fail.
Setup: Docker Compose is the standard install path. The official stack includes Postgres, Redis, machine learning container, and the main server. On a machine with 8 GB RAM, the whole thing comes up without drama. The docs are clear. Plan on about 20 minutes from zero to photos uploading.
Face and object recognition: Immich uses CLIP-based embeddings for semantic search and InsightFace for face recognition, all running locally. Semantic search (“photos of a dog at the beach”) is genuinely useful. Face clustering works well — I got accurate groupings for frequent subjects after about 5,000 photos had been processed. You merge and name clusters manually, and the UI makes that reasonably fast. The machine learning container requires enough CPU headroom or a compatible GPU. On a 4-core Intel N100-class mini PC with no GPU, indexing 40,000 photos took roughly 18 hours. Not fast, but it’s a one-time cost.
Multi-user: Full multi-user support with separate libraries. You can share albums with other users or generate public share links with optional password protection. Each user has their own timeline and storage quota controls. This is the strongest multi-user offering of the four.
Resource requirements: The ML container alone idles at around 500 MB RAM. Total stack at idle is 1.5–2 GB. During active indexing, expect 3–4 GB and near-100% CPU on all cores. Running on a Raspberry Pi 5 (8 GB) is possible but slow. A Mini PC with an N100 or better is the practical minimum for a comfortable experience.
Drawbacks: Immich hit 1.0 relatively recently, and the team still marks APIs as unstable between major versions. If you run automation around it (scripts, integrations), something will break on upgrades. The read-only partner sharing feature is good but limited — there’s no granular permission model for shared albums yet. Video transcoding is handled but requires a reasonably capable CPU or hardware acceleration setup.
Pricing: Free and open source. Hosting costs are your own hardware and electricity.
—
PhotoPrism
PhotoPrism has been around longer than Immich and has a more mature, stable codebase. It looks different — the interface has a classic web app feel rather than a mobile-first timeline — and it takes a different philosophy: it indexes photos where they live rather than importing them into a managed folder structure. If you have an existing photo archive with a folder hierarchy you’ve maintained for years, this matters a lot.
Setup: Also Docker-based. Straightforward, though the configuration options in the YAML are more extensive than Immich’s. Plan on 30–45 minutes. PhotoPrism Plus (their paid tier) unlocks some features — more on that below.
Face and object recognition: PhotoPrism uses TensorFlow under the hood. Object and scene tagging is solid and runs automatically. Face recognition is functional but the clustering UI is less intuitive than Immich’s — merging and naming faces takes more clicks, and I found the initial clustering less accurate on family photos where lighting varied a lot. For subject matter tagging (“beach,” “birthday,” “food”), PhotoPrism is very good and integrates cleanly with its search.
Multi-user: Multi-user support exists but is gated behind PhotoPrism Plus, which is a subscription. The free version is single-user. Plus pricing is around USD $4/month or roughly $6 CAD/month for personal use, or $12 USD/$16 CAD/month for business/multi-user. If you need multiple family members to have their own accounts, you’re paying for it.
Mobile backup: The PhotoPrism mobile apps are a weak point. The official mobile experience is a web app wrapped in a shell. Third-party apps like PhotoSync can push to it, but it’s nowhere near the polished first-party backup that Immich or even Google Photos offers. If automatic mobile backup is a priority, this is a real drawback.
Resource requirements: Similar to Immich when indexing, though without a separate ML container the memory overhead at idle is lower — around 800 MB to 1.2 GB for a typical setup. Indexing the same 40,000-photo library took about 22 hours on the N100 box.
Drawbacks: The mobile backup situation is the biggest daily-driver problem. The multi-user paywall will frustrate some users who expected fully open-source functionality. Development velocity has slowed compared to Immich — features that users have requested for years are still backlogged.
Pricing: Core is free and open source. PhotoPrism Plus is ~$4 USD/month or one-time license options available.
—
Nextcloud Memories
If you’re already running Nextcloud, Memories is the obvious first thing to try. It’s an app that installs inside your existing Nextcloud instance and adds a Google Photos-style timeline view over your Nextcloud Files storage. It integrates with the Nextcloud ecosystem — sharing, users, storage backends — and doesn’t require a separate stack.
Setup: Install from the Nextcloud app store. Requires the Recognize app for AI tagging and face detection. If your Nextcloud is healthy, setup is 10 minutes. If your Nextcloud is on a shared hosting setup or an older version, EXIF indexing and the required cron jobs can be fiddly.
Face and object recognition: Recognize uses locally-run neural networks for face clustering and scene tagging. Results are decent but slower than Immich’s dedicated ML container on the same hardware. Face naming is workable. Object tags are useful for search. Performance depends heavily on how beefy your Nextcloud server is, since there’s no offloading to a separate container.
Multi-user: This is where Nextcloud Memories genuinely leads. If you’re already using Nextcloud for family or team file sharing, your user accounts and permissions carry over automatically. Each user gets their own Memories timeline over their own file space. Sharing albums or individual photos uses Nextcloud’s existing sharing model, which is mature and flexible. No additional subscription required.
Mobile backup: The Nextcloud mobile apps include auto-upload for photos. It works reliably on Android; the iOS version has historically been flakier with background uploads but has improved. It’s not as slick as Immich’s dedicated photo app but it’s functional.
Resource requirements: You’re adding load to an existing Nextcloud instance. Recognize’s indexing is CPU-intensive. On a low-power box already running Nextcloud, you’ll feel it. Not recommended for a Nextcloud instance on less than 4 GB RAM if you want a smooth experience during indexing.
Drawbacks: Memories is only as good as your Nextcloud setup. If your Nextcloud has performance problems, Memories inherits all of them. The UI, while improved significantly in recent versions, still feels like an app inside a platform rather than a purpose-built photo manager. Video playback and transcoding are more limited than Immich. If you’re not already on Nextcloud, setting it up just for Memories is the wrong approach — use Immich or PhotoPrism instead.
Pricing: Free. Both Nextcloud and Memories are open source.
—
LibrePhotos
LibrePhotos is the underdog here, and it deserves an honest look even if it’s not the first recommendation for most people. It’s a fully open-source Google Photos clone — timeline, map view, face recognition, albums, sharing. The demo looks great.
Setup: Docker Compose. More containers than Immich (it includes its own worker, backend, frontend, and Postgres). Setup is workable but the documentation is patchier than the others. Budget 45–60 minutes and expect to troubleshoot at least one thing.
Face and object recognition: LibrePhotos uses dlib for face detection and recognition. The accuracy is noticeably lower than Immich’s InsightFace implementation on diverse lighting conditions and partial faces. Object/scene tagging uses PlacesCNN for location scenes. It works, but the search experience isn’t as smooth as PhotoPrism’s tagging or Immich’s CLIP semantic search.
Mobile backup: No official first-party mobile app. You’re relying on third-party tools or manual uploads. For a daily-driver replacement, this is a serious gap.
Multi-user: Multi-user is supported and free. Each user has their own photo library. Sharing albums between users works. It’s not as polished as Immich’s implementation but it functions.
Resource requirements: Lighter than Immich at idle, around 800 MB–1.2 GB. Indexing speed is similar to PhotoPrism.
Drawbacks: Development activity has been inconsistent. There have been periods of months with no commits, followed by bursts of activity. In a category where mobile OS changes can break backup apps overnight, that’s a real risk. The mobile backup gap alone rules it out as a primary solution for most households. It’s most useful as a read/browse interface for an archive you manage another way.
Pricing: Free and open source.
—
Quick Comparison
| Feature | Immich | PhotoPrism | Nextcloud Memories | LibrePhotos | |—|—|—|—|—| | Mobile auto-backup | Excellent | Poor (no first-party app) | Good | None | | Face recognition quality | High | Medium | Medium | Lower | | Semantic search | Yes (CLIP) | No | No | No | | Multi-user (free) | Yes | No (paid) | Yes | Yes | | Setup difficulty | Easy | Easy | Easy (if on NC) | Moderate | | RAM at idle | ~1.5–2 GB | ~800 MB–1.2 GB | Adds to existing | ~800 MB–1.2 GB | | Active development | Very active | Moderate | Active | Inconsistent | | Folder-first indexing | No | Yes | Yes (via NC Files) | No |
—
Conclusion: Which One to Pick
For most people making the switch from Google Photos: Use Immich. The mobile app is reliable, face recognition is the best of the group, semantic search is actually useful, and multi-user support doesn’t cost extra. The resource requirements are real — you need at least an N100-class box with 8 GB RAM for a comfortable experience — but that’s a ~$150–200 CAD investment in a used mini PC that will also run your other homelab services.
If you have an existing photo archive in a folder hierarchy you want to preserve: PhotoPrism handles this better than Immich’s import model. Just go in knowing the mobile backup situation is poor and multi-user will cost you.
If you’re already running Nextcloud: Try Nextcloud Memories before setting up a separate stack. If it meets your needs, great — you’ve saved yourself another service to maintain. If the performance on your hardware is frustrating or the feature gaps bother you, then add Immich alongside.
LibrePhotos: Hard to recommend as a primary solution right now given the mobile backup gap and development consistency concerns. Keep an eye on it — the concept is solid — but it’s not the right daily driver for 2026.
The honest summary: this category has come a long way. Immich in particular has closed most of the practical gap with Google Photos for home use. The remaining differences — video transcoding quality, shared album collaboration for outside users, automated memory/highlight reels — are real but liveable for most people who just want their photos under their own control.
Related Auburn AI Products
Building a homelab or self-hosting content site? Auburn AI has practical kits:
- 500 Homelab and Self-Hosting Blog Titles ($27)
- Auburn AI Monitoring Stack ($37) – 6 production PowerShell scripts
- Podcast Automation Kit ($37)
- Browse all Auburn AI products