Own your photo library β no iCloud, no Google Photos, no monthly fee. This guide covers picking the right app, the hardware you need, migrating off cloud, and the mistakes that bite you later. My own library is about 54,000 photos and videos (roughly 920 GB) on the setup described further down.
Immich vs Synology Photos: the short answer
Choose Immich if you want a genuine Google Photos replacement: reliable background auto-backup, accurate face recognition, CLIP-based semantic search and hardware acceleration on NVIDIA (CUDA), AMD (ROCm) or Intel (OpenVINO) graphics. The trade-off is a ~30-minute Docker setup and ongoing maintenance you own.
Choose Synology Photos if you already have a Synology NAS and want photos working in about three minutes with near-zero upkeep. You give up semantic search and settle for basic face recognition and a slower mobile app.
| If you… | Better fit | Why |
|---|---|---|
| Want the closest thing to Google Photos | Immich | Reliable auto-backup + semantic search |
| Already own a Synology NAS | Synology Photos | Native integration, ~3-minute setup |
| Want the lowest maintenance | Synology Photos | Appliance-like; no containers to manage |
| Want to search photos by what is in them | Immich | CLIP semantic search; Synology is subject/place only |
| Rely on a good phone app for the whole family | Immich | Excellent iOS + Android apps, shared albums |
| Have an NVIDIA, AMD or Intel GPU to use | Immich | Broadest hardware-acceleration support |
| Want stability and a clean UI over features | PhotoPrism | Simpler scope; manual backup though |
π― Why Self-Host Photos in 2026
The trade-off: you’re your own IT department. Backups, hardware, updates. If you already run a homelab or NAS, the overhead is minimal β ~2 hours setup, ~1 hour/quarter maintenance.
π Immich vs PhotoPrism vs Synology Photos
| Immich | PhotoPrism | Synology Photos | |
|---|---|---|---|
| Best for | Google Photos replacement | Stability + clean UI | Zero-config Synology owners |
| π± Mobile app | Excellent (iOS + Android) | Decent (PWA) | Functional but slow |
| βοΈ Auto backup | β Reliable background | β Manual / syncthing | β Native integration |
| π€ Face recognition | β Accurate | β With retraining | β οΈ Basic |
| π Semantic search | β CLIP-based | β οΈ Limited tags | β Subject + place only |
| π¨βπ©βπ§βπ¦ Multi-user | β Shared albums | β οΈ Limited | β Yes |
| β‘ HW acceleration | NVIDIA (CUDA), AMD (ROCm), Intel (OpenVINO) | NVIDIA only | Synology models only |
| β±οΈ Setup time | ~30 min (Docker) | ~20 min (Docker) | ~3 min (App Store) |
π₯οΈ Hardware You Need
Storage Sizing
Rough math: 2-4 MB per photo, 50-500 MB per minute of video. Family of 4 with iPhones = 60-120 GB/year.
ML Acceleration Options
-
π’
NVIDIA GPU (GTX 1660+) Fastest option
Required for serious CLIP semantic search. -
π
Apple Silicon (Mac Mini/Studio) Silent, low-power
Metal Performance Shaders accelerate Immich’s ML well.
π Migration Paths
βοΈ My Setup
Two-tier: Synology DS723+ as storage (two 8 TB drives mirrored, 7 TB usable) + an Intel NUC 8 running Immich in Docker. There is no GPU: the machine-learning jobs (face recognition and search) run on the CPU. Storage and compute scale independently.
Backups: this is the gap in my setup. The mirror protects against one drive failing and Immich dumps its database every night, but I do not have an offsite copy yet. The 3-2-1 rule below is what I should be doing, not what I do today.
β οΈ Common Mistakes
- β Consumer drives in 24/7 NAS β they die fast. NAS-rated (WD Red Plus) costs one Google One month more and lasts years longer.
- β RAID 0 “because I have backups” β when a drive dies you spend a weekend restoring. RAID 1 mirror saves that hassle.
- β No offsite backup β RAID is not a backup. Fire, theft, ransomware can take everything.
- β Everything on a Raspberry Pi β Pi 5 can serve photos but ML processing crawls. Offload ML to a more capable box.
- β All eggs in Synology Photos β locks you in. Use Immich on top of Synology storage instead.
π Where to Go Next
πΈ Immich Without a GPU (Timed)
Search and face indexing timed on an Intel NUC: about 51,000 items in under an hour.
π Homelab Hardware Guide
Tested picks for NAS, GPUs, accelerators, and mini PCs.
π― LLM Hardware Checker
Same homelab can run local AI. Check if your hardware is up to it.
Build This Stack
Affiliate links β helps support testing at no extra cost to you.
Frequently Asked Questions
What is the best self-hosted photo library in 2026?
Immich. Best mobile app, most active development, closest to Google Photos. PhotoPrism is runner-up for stability.
How much storage do I need?
~1GB per 200-400 photos. Family of 4: plan 2-4TB for 5 years. Synology DS224+ with two 4TB WD Red Plus in RAID 1 = 4TB usable.
Do I need a GPU or Coral TPU?
Not required but much faster. A CPU is slower but enough for most libraries: I timed about 51,000 items in under an hour on an Intel NUC. Immich does not support the Google Coral TPU β a GPU (NVIDIA CUDA / AMD ROCm) or Intel iGPU (OpenVINO) is what speeds up semantic search and face processing.
Is Immich a true Google Photos replacement?
Yes for most users. Timeline, albums, faces, search, mobile backup, sharing β all work.
How do I migrate from iCloud or Google Photos?
Google: Takeout + Immich CLI. iCloud: icloudpd. Amazon/Dropbox/OneDrive: desktop sync β Immich library importer.
What about backups?
3-2-1 rule: three copies, two media types, one offsite. RAID protects against drive failure, not deletion or fire.