Immich: The Complete Guide to Self-Hosting Your Own Google Photos in 2026

Privacy-first photo management that your whole family can actually use.

Let’s be honest about Google Photos.

It’s convenient. It works. Your photos sync automatically. The AI search is creepily good at finding “that photo from the beach trip in 2019.” And for years, that was enough.

But then you started paying attention.

You noticed the “15GB free storage” that somehow fills up every year. You read about Google using your photos to train AI. You wondered what happens to your memories when policies change. And you realized: your photos aren’t really yours anymore.

Enter Immich.

In 2026, Immich has become the go-to solution for self-hosted photo management. With 87,000+ GitHub stars, an active development team, and feature parity with Google Photos that actually exceeds it in some areas, it’s no wonder homelab enthusiasts are making the switch.

This guide covers everything you need to know to set up Immich, configure mobile backup, enable AI features, and migrate your existing photo library—all while keeping your memories private on your own hardware.

Why Immich? The Case for Self-Hosted Photos

The Privacy Argument

Every photo you upload to Google Photos, iCloud, or Amazon Photos lives on someone else’s servers. These companies have access to your memories:

Photos of your kids growing up
Location data from every vacation
Faces of everyone you know, tagged and searchable
Documents, receipts, and sensitive information you’ve captured

When you self-host with Immich, all of that stays on your hardware. No cloud provider can train models on your family photos. No policy change can lock you out. No subscription fee can hold your memories hostage.

The Storage Argument

Google offers 15GB for free. After that, you’re paying monthly for storage—forever.

With Immich, your only limit is your hard drive. A 4TB drive costs $80 once. The equivalent Google One storage? $150/year, every year, indefinitely.

The Feature Argument

Here’s what might surprise you: Immich has more features than Google Photos in several areas:

Feature	Immich	Google Photos
Face Recognition	✅ Local	✅ Cloud
Smart Search	✅ Local	✅ Cloud
Automatic Backup	✅	✅
HDR Video Support	✅	✅
Memories/Timeline	✅	✅
Album Sharing	✅ Custom URLs	✅
Duplicate Detection	✅ User-controlled	❌
Manual Face Tagging	✅	❌
External Library Import	✅	❌
Self-hosted	✅	❌
No Subscription	✅	❌

The “Family Test”

The real proof? My family actually uses it.

That’s the difference between a pet project and a genuine Google Photos replacement. Immich’s interface is familiar enough that non-technical users can pick it up in minutes. The mobile app works the same way. Automatic backup just… works.

Hardware Requirements

Before diving into setup, let’s talk about what you need to run Immich.

Minimum Requirements

For a functional Immich server with machine learning features:

Component	Minimum	Recommended
RAM	6GB	8GB+
CPU	2 cores	4+ cores
Storage	Depends on library	SSD for database
OS	Linux/Windows (WSL2)	Linux

The 6GB RAM requirement is real—machine learning models for face recognition and smart search need memory. If you’re running on less (like a Raspberry Pi with 4GB), you can offload ML to a remote server or use lighter models.

GPU Acceleration (Optional but Recommended)

If you want snappy face recognition and instant smart search, GPU acceleration makes a massive difference. Immich supports:

:::note[Supported GPU Backends]

CUDA: NVIDIA GPUs (compute capability 5.2+) — most reliable
ROCm: AMD GPUs
OpenVINO: Intel GPUs (Iris Xe, Arc)
ARM NN: Mali GPUs
RKNN: Rockchip SoCs (RK3566, RK3568, RK3576, RK3588) :::

CPU-only works fine for small libraries. But once you hit 10,000+ photos, you’ll appreciate the GPU speedup.

Installation: Docker Compose Setup

The fastest way to get Immich running is with Docker Compose. Here’s a complete walkthrough.

Step 1: Create the Immich Directory

mkdir -p ~/immich-app
cd ~/immich-app

Step 2: Download Configuration Files

# Download the docker-compose.yml
wget -O docker-compose.yml https://github.com/immich-app/immich/releases/latest/download/docker-compose.yml

# Download the example environment file
wget -O .env https://github.com/immich-app/immich/releases/latest/download/example.env

Step 3: Configure Environment Variables

Edit the .env file with your settings:

# The location where your uploaded files are stored
# Point this to your storage location (NAS mount, large drive, etc.)
UPLOAD_LOCATION=/mnt/photos-library

# Database location - keep this local, not on network storage
DB_DATA_LOCATION=./postgres

# Set your timezone
TZ=America/New_York

# Immich version - v2 is the current stable release
IMMICH_VERSION=v2

# Database password - CHANGE THIS to something secure
DB_PASSWORD=your-secure-random-password-here

# These don't need to change
DB_USERNAME=postgres
DB_DATABASE_NAME=immich

:::warning[Important: Database Location] Never put DB_DATA_LOCATION on a network share (NFS, SMB). PostgreSQL requires local filesystem access. Your photos can go anywhere, but keep the database local. :::

Step 4: Start the Containers

docker compose up -d

This pulls and starts:

immich-server: Main application
immich-machine-learning: AI features (face recognition, smart search)
immich-redis: Caching and job queue
immich-postgres: Metadata database

Step 5: Access the Web UI

Navigate to http://<your-server-ip>:2283 and click “Getting Started.”

The first user to register becomes the admin. This is your admin account, so use a strong password.

Mobile App Setup

Immich shines with its mobile apps—available on iOS, Android, and F-Droid.

Download Options

Google Play Store: Search “Immich”
Apple App Store: Search “Immich”
F-Droid: For the privacy-pure option
GitHub Releases: Direct APK download
Obtainium: Auto-update directly from GitHub

First-Time Setup

Open the app
Enter your server URL: http://<your-server-ip>:2283
Log in with your admin credentials
Grant photo library access when prompted

Configuring Automatic Backup

Tap the cloud icon in the top-right to access backup settings:

Backup Options:
  - Select specific albums or "All Photos"
  - Choose Wi-Fi only or allow cellular
  - Enable background backup
  - Set backup frequency

:::tip[Pro Tip: Initial Upload] If you have thousands of photos, the initial backup will take time. Start with a small album to test, then enable full backup overnight. You can monitor progress in the server’s Job Status. :::

AI Features: Face Recognition and Smart Search

This is where Immich genuinely competes with the big players.

Face Recognition

Immich automatically detects faces in your photos and groups them by person:

Navigate to People in the sidebar
Review detected faces
Assign names to people
Immich learns and improves over time

You can also manually tag faces—useful when the AI misses someone or groups incorrectly.

Smart Search

This is the “find photos of dogs at sunset” feature, and it works remarkably well:

Search examples that work:
  - "beach sunset"
  - "birthday party"
  - "my dog"
  - "screenshot"
  - "document"

Behind the scenes, Immich uses CLIP models to understand image content semantically. The magic? All processing happens locally on your server.

Duplicate Detection

Immich finds similar photos and lets you decide what to keep:

Go to Utilities > Duplicates
Review detected duplicates
Keep or delete each group

Unlike Google Photos which might auto-delete, Immich asks for your approval first. No accidental data loss.

GPU Acceleration for Machine Learning

If you have a GPU, enabling hardware acceleration significantly speeds up ML tasks.

NVIDIA (CUDA)

For NVIDIA GPUs with compute capability 5.2+:

# Download hardware acceleration config
wget -O hwaccel.ml.yml https://github.com/immich-app/immich/releases/latest/download/hwaccel.ml.yml

Then edit your docker-compose.yml:

immich-machine-learning:
  container_name: immich_machine_learning
  image: ghcr.io/immich-app/immich-machine-learning:${IMMICH_VERSION:-release}-cuda
  extends:
    file: hwaccel.ml.yml
    service: cuda
  volumes:
    - model-cache:/cache
  env_file:
    - .env

:::note[Prerequisites]

NVIDIA driver 545+ (CUDA 12.3 support)
NVIDIA Container Toolkit installed :::

Intel (OpenVINO)

For Intel integrated or discrete GPUs:

immich-machine-learning:
  image: ghcr.io/immich-app/immich-machine-learning:${IMMICH_VERSION:-release}-openvino
  extends:
    file: hwaccel.ml.yml
    service: openvino

Raspberry Pi / Rockchip (RKNN)

For ARM boards like Orange Pi or Radxa:

immich-machine-learning:
  image: ghcr.io/immich-app/immich-machine-learning:${IMMICH_VERSION:-release}-rknn
  extends:
    file: hwaccel.ml.yml
    service: rknn

Confirm It’s Working

Check GPU utilization during face detection jobs:

# NVIDIA
nvidia-smi

# Intel
intel_gpu_top

# AMD
radeontop

Or check the ML container logs:

docker logs immich_machine_learning | grep -i cuda
# Should show: Available ORT providers: ['CUDAExecutionProvider', ...]

Migrating from Google Photos

If you’re switching from Google Photos, here’s how to bring your library with you.

Option 1: Google Takeout + immich-go

Step 1: Export from Google

Go to Google Takeout
Select only “Photos”
Choose delivery method and format
Wait for the export (can take days for large libraries)

Step 2: Import with immich-go

immich-go is a specialized tool for Google Takeout imports:

# Download immich-go
wget https://github.com/simulot/immich-go/releases/download/v0.24.0/immich-go_Linux_x86_64.tar.gz
tar -xzf immich-go_Linux_x86_64.tar.gz

# Run the import
./immich-go upload \
  --server http://your-server:2283 \
  --api-key YOUR_API_KEY \
  --from /path/to/takeout

This preserves albums, dates, and metadata from your Google Photos library.

Option 2: Direct Upload via Web UI

For smaller libraries:

In Immich web UI, go to Upload
Drag and drop photos/videos
Use folder structure to organize

Option 3: External Library

If your photos are already organized on a drive:

Go to Administration > Libraries
Create an external library
Point to your existing photo directory

Immich will scan and import without duplicating files.

Backup Strategy

Self-hosting means you’re responsible for backups. Here’s how to protect your memories.

What to Back Up

Component	Backup Method
Photos/Videos (`UPLOAD_LOCATION`)	File-level backup to remote storage
Database (`DB_DATA_LOCATION`)	Immich’s built-in backup or pg_dump
`.env` file	Keep a copy in a password manager

Immich’s Built-In Database Backup

Immich automatically backs up the database. Configure in Administration > Settings > Backup:

Backup frequency
Retention period
Storage location

:::warning[Crucial Reminder] The database contains metadata only (albums, tags, face data). Your actual photos live in UPLOAD_LOCATION. You must back up both independently. :::

Recommended Backup Setup

# Example: Daily backup to a remote NAS
0 3 * * * rsync -av /mnt/photos-library/ backup-nas:/backups/immich-photos/
0 4 * * * docker exec immich_postgres pg_dump -U postgres immich > /backups/immich-db-$(date +\%Y\%m\%d).sql

What’s Coming in 2026

The Immich team has ambitious plans. Here’s what’s on the roadmap:

Near-Term Features

Built-in image editor (most requested!)
Workflows for automation
Database restore from web UI
Integrity check tools

2025 Achievements (Context for What’s Next)

The team shipped 85 new features in 2025:

HDR video support
Search by tags and descriptions
Manual face tagging
Persistent memories
OCR for text in images
Google Cast support
Private/locked photos
“View similar photos” discovery

Version 2.0 (October 2025) was a major milestone—a complete rewrite of the sync infrastructure and mobile app, making the experience butter-smooth.

Platform Compatibility

Immich runs on almost anything:

Platform	Support Level
Docker	✅ Primary method
TrueNAS	✅ Built-in container app
Unraid	✅ Community template
Proxmox	✅ LXC/VM
CasaOS	✅ App store
Kubernetes	✅ Helm charts available
Raspberry Pi	✅ ARM supported
Synology	⚠️ Docker required

Troubleshooting Common Issues

Photos Not Uploading from Mobile

Check server URL in app settings
Verify background backup is enabled
Ensure app has photo library permissions
Check server logs: docker logs immich_server

Face Recognition Not Working

Verify ML container is running: docker ps | grep immich
Check ML logs: docker logs immich_machine_learning
Trigger manual scan: Administration > Jobs > Face Detection

Slow Performance

Low RAM: Disable some ML features or use lighter models
CPU bottleneck: Enable GPU acceleration
Database on network storage: Move to local SSD
Large library: Increase job concurrency in settings

Can’t Access Web UI

Check container status: docker compose ps
Verify port 2283 isn’t blocked by firewall
Check for port conflicts: sudo lsof -i :2283

The Bottom Line

Immich represents a genuine paradigm shift in personal photo management.

For years, we’ve traded privacy for convenience—uploading our most personal memories to servers we don’t control, accepting subscription fees for our own data, and hoping companies don’t change their policies.

Immich flips that equation. You get:

Complete privacy — Your photos never leave your network
No subscriptions — Your storage is your own
Full control — Customize, extend, and own your setup
Feature parity — Everything Google Photos offers, locally

And with the v2.0 release in late 2025, Immich has reached the maturity where your family will actually use it. That’s the real test, and Immich passes.

If you have a homelab—even a small one—Immich deserves a spot in your setup. Your memories are worth it.

Quick Reference

Important URLs

Documentation: https://docs.immich.app/
GitHub: https://github.com/immich-app/immich
Community: https://discord.gg/immich
immich-go (migration tool): https://github.com/simulot/immich-go

Port Reference

Web UI: 2283
ML Service: 3003 (internal)

Useful Commands

# View logs
docker compose logs -f

# Restart services
docker compose restart

# Update to latest
docker compose pull && docker compose up -d

# Check GPU usage
nvidia-smi -l 1

Happy self-hosting. Your photos, your rules.