Faces You Know, Photos You Can Find: How Immich's Facial Recognition Works

We all have the same problem: you want to find all the photos of your daughter’s first birthday party, or every picture your partner appears in, or how your mum looked ten years ago. Scrolling through hundreds or thousands of photos one by one is exhausting. Immich solves this with facial recognition, a powerful feature that groups people together and lets you search by who is in the photo. The catch? It does all of this entirely on your own server, keeping your biometric data exactly where it belongs.

How Immich Recognises Faces

Immich uses machine learning to analyse the photos in your library and detect faces. When it finds a face, it doesn’t store a picture of it. Instead, it creates a mathematical representation called an embedding, a numerical fingerprint that captures the unique characteristics of that person’s face. These embeddings are then compared with each other to group similar faces together.

The key here is that all of this happens on your own hardware. You upload your photos to your Immich server, and the machine learning runs right there. No image data, no face data, no embeddings leave your server. No Google, no Apple, no cloud company in the world sees these photos. This is how privacy-respecting photo management actually works.

Meet Your People

Once Immich has detected and grouped faces, it creates a “People” section in your library. You’ll see clusters of faces that the algorithm believes belong to the same person. At first, these are unnamed. Your job is simple: name them.

Click on a cluster, and Immich will show you every photo where that person appears. Give them a name, and you’ve created a searchable person. Do this for everyone in your library, and suddenly you have a powerful tool: search your photos by name.

You can also set a cover photo for each person, which shows up in your People grid. Pick a photo where their face is clear and well-lit so you can quickly spot them. You might want to mark your closest family and friends as favorites, and they will appear at the top of your People list.

Refining the Recognition

Facial recognition isn’t perfect. Sometimes Immich will split one person into multiple clusters because the lighting was different, they were at a distance, or they wore glasses. Sometimes it will group two different people together. Both of these problems are easy to fix.

If one person appears in multiple clusters, open both clusters, select photos from each, and merge them. Immich updates the embeddings and learns from your correction. If two people are wrongly grouped together, you can move individual photos to a new cluster or simply split the group. Over time, as you refine these clusters, the accuracy improves.

You can also delete incorrect detections. If a photo has a false positive (Immich thought the sunset or an animal was a face), remove it.

Why This Matters in Europe

In Europe, we take privacy seriously. The GDPR and other regulations exist because biometric data is sensitive. Companies like Google and Apple use your photos and faces to train their algorithms, to improve their services, and to build advertising profiles. They don’t always ask explicit permission, and your face data often ends up shared across multiple services.

With Immich, your biometric data never leaves your server. You own it. You control it. There is no training algorithm sucking up your faces. There is no sharing, no selling, no profiling. Facial recognition in Immich is a tool for you to organise your own photos, not a surveillance mechanism built into someone else’s infrastructure.

Better Results with Better Input

If you want the best results from facial recognition, keep a few things in mind. Clear, front-facing photos work better than blurry ones or extreme angles. A library with many photos of the same people will produce more accurate groupings than a library with only a few photos per person. Dedicated portrait photos are gold: they let the algorithm learn the unique characteristics of each face.

If you have a brand new Immich instance with thousands of photos, processing will take some time. Facial recognition is computationally intensive. Depending on your hardware, analysing a large library can take hours or even days. But it runs in the background, and you can keep using Immich while it works.

Find Photos in Seconds

Once you’ve named your people, searching becomes effortless. Click on a person’s name and see every photo they appear in. Scroll through your favourite moments with them. Create collections of memories. Share photos with family by person, not by random dates and folders.

This is what a privacy-respecting photo storage service looks like. Powerful tools for organising your own memories, running entirely on your own hardware, no surveillance, no tracking, no third parties.

Ready to try Immich with facial recognition? PixelUnion makes it simple. We handle the setup, the hardware, and the privacy so you can focus on your photos. Visit https://pixelunion.eu to get started.