Tool profile
Immich: cost, fit and limits
Immich is an open source, self-hosted photo and video backup server licensed under GNU AGPL v3 that detects and groups faces through a machine learning service and offers web, iOS and Android clients for backing up phone photos and videos.
What to know first
Immich is free software that you install with Docker Compose on your own machine. The documentation we could read describes facial recognition that groups people through DBSCAN clustering, with settings for model, detection score and recognition distance. The smart search page was not readable, so CLIP search is not claimed here. The quick start asks for 6 GB of RAM and 2 CPU cores, and no company is named as maintainer.
Immich at a glance
| Maker | Not named on the vendor pages, Not stated on the vendor pages |
|---|---|
| Free plan | Yes |
| Entry price | Quote only |
| API access | No |
| Category | Best AI Photo Organizer Software in 2026, rank 4 of 12 |
| Website | immich.app |
Who Immich is best for
Households replacing a cloud photo app
Android and iOS clients plus a web app mean phones can back up to a home server, which is the main reason people pick Immich over desktop catalogs.
Tinkerers who tune recognition
Face clustering exposes model, detection score and recognition distance settings, so someone who is unhappy with merged or split people can adjust thresholds.
People with a spare home server
A Docker Compose quick start and a Job Queues tab for watching processing suit owners who already run containers and monitor them.
Immich plans and pricing
| Plan | Price | What it includes |
|---|---|---|
| Open source (self-hosted) | Free under AGPL v3 | Server with a Docker Compose quick start, web app, Android and iOS clients, and an APK through GitHub Releases. |
| Support the project | No price published | Donation or purchase links and a merchandise store; no plan terms shown on the opened pages. |
| Mobile and web clients | Included, no separate price listed | Web app, Android and iOS clients that connect to the server you host, with an APK on GitHub Releases for Android. |
Checked November 2026.
What Immich does
- Facial recognition
- Faces in photos and videos are detected and grouped into people, which you can name and search by, through a separate machine learning service that runs next to the main server.
- DBSCAN clustering
- The grouping is based on DBSCAN, and the documentation lists configurable model, detection score and recognition distance for tuning results when people are merged or split wrongly.
- Optional acceleration
- The machine learning service supports optional hardware acceleration, so a machine with a compatible GPU can process large libraries faster than a plain processor would manage.
- Job queues
- A Job Queues tab lets you watch the server work through photos, which shows whether a big import has finished, is still running or is stuck on a failing step.
- Mobile and web clients
- Clients exist for the web, iOS and Android, with an APK available via GitHub Releases and mentions of Obtainium and FUTO F-Droid as ways to install it on phones without the Play Store.
How Immich handles your data
The opened documentation is self-hosted software documentation. It says the server is installed with Docker Compose and the user picks the host, which means photos stay wherever the owner runs it. Retention is not described, and no training use of customer data, hosting region or certification is stated on the pages read. The site names no operating company as the maintainer, although FUTO is linked as a company without a statement that it maintains Immich. The vendor's policy was not checked for anything beyond the documentation pages.
Where Immich falls short in practice
- Hardware needs. The quick start lists 6 GB of RAM and 2 CPU cores as a floor, which excludes the smallest single-board computers and older network storage boxes.
- You own the backups. Docker Compose and a self-run server mean storage, updates and backup copies are the owner's job, and the pages we read describe no managed hosting option.
- Unclear steward. No maintainer company is named on the pages read, so a buyer wanting a support contract or a named data controller has nobody to ask.
Works with
Alternatives to Immich
Immich: common questions
What hardware does Immich need?
The quick start lists at least 6 GB of RAM and 2 CPU cores. The machine learning service supports optional hardware acceleration, so a GPU can help with large libraries, though the pages we read did not give a minimum disk size.
Does Immich have a paid plan?
The pages read show the software as open source under GNU AGPL v3 and link a Buy Immich store and a Support the Project page, with no prices or plan tiers visible. Treat it as free software with optional donations.
How does Immich group faces?
Face clustering is based on DBSCAN, and the docs list settings for the model, the detection score and the recognition distance. People can then be named and searched. Tune the recognition distance if people are merged or split wrongly, then re-run the job.
Can I use Immich from my phone?
Yes. Clients exist for the web, iOS and Android, with an APK on GitHub Releases and mentions of the App Store, Google Play, Obtainium and FUTO F-Droid. The phone talks to the server you host, so it needs network access to it.
Who maintains Immich?
The pages read name no maintaining company. FUTO is linked as a company, but the page does not say it maintains Immich. A buyer who needs a named vendor for contracts or data processing agreements should ask the project directly before relying on it.