Skip to main content

FAQ

If something is broken, start with Troubleshooting.

Before you install​

Does it change my Immich library?

Ordinary selection reads your library. When delivery is enabled (--upload-to-immich, upload.enabled, or the web UI's upload switch), it uploads the finished film, creates the immich-memories/generated tag if needed and tags the asset. It can also create the requested album and add the film to it. A re-render into the same album moves the app's superseded copy to Immich's trash; it does not hard-delete it. What Immich sees lists these writes.

What leaves my network?

Library processing talks to your Immich server. Model setup downloads weights from their distribution hosts. Optional readers, caption servers, translated place names and map fly-overs can contact other services; Privacy lists the requests and what they send.

Will it run on my NAS?

Yes, it works on a plain NAS: the default install cuts films on a NAS CPU from dates, places, favourites, people and what small local classifiers measure on each picture. A GPU makes it faster and adds captions, and a text model can refine the draft (What a GPU or a model adds). Sizes and the one Synology trap are on On a NAS and Requirements.

Do I need face recognition?

No. Without a person, a period covers everyone. Named faces make people films possible and let it keep close family in the film (Home and people).

iPhone only?

Anything FFmpeg decodes. Live Photos are tested on iPhones; Samsung and Pixel motion photos are untested.

After the first film​

Why is this picture in (or not)?

immich-memories runs why <asset id> says where it passed or was dropped and why. The rules are on How it chooses.

Can I pick pictures myself?

Yes. Tick or untick in the web UI’s pool and select Preview with these choices, star it in Immich, or pass --include / --exclude. A tick outranks the editor. See Edit the cut.

Why is the first cut slow and the second fast?

The first cut measures each picture it can reach once and banks the result; the second is mostly the render. prepare does a period ahead of time, overnight if you like.

How much disk?

Caches are capped by config: 10 GB of downloaded video (kept 7 days), 10 GB of Immich previews, plus the store (store.db). Films come on top, sized by length, resolution and codec.

Can it make films on its own?

Yes. The default automation.cooldown_hours: 24 suits a daily schedule; it is configurable, and auto run --force bypasses that cooldown. Rotation rules still apply. Automatic films covers setup.

Several people on one Immich server?

The web UI is single-user, single-replica: one primary Immich account (the upload target). Run one instance per library. People whose pictures use separate accounts can still get one people film from both: add the second account under immich.accounts and run generate --accounts primary,partner on the CLI (A second Immich account); automation.accounts does the same for the daily run, and Immich accounts to read on the web UI's New memory page for a film made there.

Is it stable?

Install, read-only Immich access and rendering are. Selection keeps improving, measured on real libraries, so review a cut before you show it at a family party. The project is built with AI on purpose, as an experiment in keeping a complex codebase clean that way: Why this exists.