Immich v3: the Workflows era
Immich's v3 line (through v3.4, November 2026) turned the best self-hosted photo platform into something more ambitious: Workflows — an event-driven automation system inside Immich — plus JPEG XL support that can shrink libraries dramatically, and a steady drumbeat of editor and mobile improvements. If you run Immich from our setup guide, this is your upgrade-and-enable tour; if you do not run it yet, this is what you are signing up for in 2026.
Published Nov 18, 2026
Workflows: Immich grows an automation engine
The pattern is familiar to any Home Assistant user: a trigger fires, filters decide, actions run — assembled in a drag-and-drop builder under Utilities → Workflows, with premade templates to start from. Precision matters here, because Workflows is still a preview feature with a deliberately small set of building blocks. The triggers are asset events: an asset is created (uploaded), and metadata extraction finishing — v3.4 added a third, for OCR completion. The filters are file name or upload path, asset type, EXIF metadata, date, and location. The actions are: add to album(s), add tags, archive, set visibility or move to the locked folder, favourite — and a webhook, the escape hatch that hands off to your own scripts for anything beyond that list.
So the honest starter recipes are the ones those blocks actually build. The Immich team's own demo: an upload whose file name matches your phone's screenshot pattern → add it to a Screenshots album and archive it out of the main timeline. Two popular wishes are not built in yet, whatever forum posts imply: there is no face-recognized trigger — the team has said publicly that one is in development, precisely to enable person-based smart albums — and no filter selects assets by who uploaded them, so a partner's uploads cannot yet be auto-routed into the shared family album by a workflow. Transcoding likewise stays a background-job policy, not a workflow action. Workflows productize some of what the community used to bolt on (the old immich-autotag and friends); the webhook action is Immich acknowledging the rest still needs scripts. Start with two or three rules the blocks genuinely support, and resist building forty on day one; automation debt is real in every platform.
JPEG XL: the storage conversation
v3.4's JPEG XL support matters because photo libraries are the NAS's growth stock: the server can now generate previews in JPEG XL (it joined the format settings as a target format), and the mobile apps decode the format natively. JPEG XL encodes the same visual quality at meaningfully smaller sizes than the JPEG previews Immich generates — across tens of thousands of photos, that compounds into real headroom on the drive. The honest caveats: it applies to generated assets only — your originals are never touched (Immich's founding promise), so this does not shrink the original library itself; device and browser support for the format outside Immich's own apps is broad but not universal; and re-encoding a large existing library is a background job measured in evenings, not minutes. Enable it for new uploads, and let the migration job chew through history at night.
Measure your own saving before believing anyone's percentage — including ours. The saving depends entirely on your settings and your photos, and it is easy to measure properly:
- Write down your current image settings: preview format, preview resolution (1440px unless you changed it) and the quality value. Those three numbers define what a 'preview' even is in the comparison.
- Pick a representative sample — one album of a few hundred photos from your usual cameras. Note the total size of those assets' generated preview files (they live in the thumbs folder of your Immich data directory), or check a sample of individual preview files and average them.
- Switch the preview format to JPEG XL — same resolution, same quality — and re-run thumbnail generation for just that sample from the admin jobs panel.
- Measure the same assets again. Per-photo saving × your library's photo count is your library's saving. No extrapolation from a stranger's library required.
One worked example, explicitly illustrative — an arithmetic demonstration with stated assumptions, not a measured benchmark: suppose a 1440px preview at quality 80 averages 480 KB as JPEG and 320 KB as JPEG XL. Across a 30,000-photo library, previews fall from about 14.4 GB to about 9.6 GB — roughly 5 GB handed back. Worth having; not terabytes, and libraries with higher preview resolutions will see bigger absolute numbers. Your figures will differ from these invented ones, which is the entire point of the four steps above.
The rest of the v3 haul
- Editor upgrades — the built-in editor keeps absorbing the quick-fix jobs (crop/rotate/adjust) that used to mean a round-trip to another app.
- Mobile tagging & management — v3.4's mobile improvements make the phone app a real curation tool: tag, album and organize from the couch instead of deferring to a desktop session that never happens.
- Performance groundwork — the v3 series' database and API work shows up as a generally snappier large-library experience; our hardware acceleration guide remains the other half of that story.
Upgrading from v2 (or earlier v3)
- Back up the Immich database and your docker-compose + .env — the standard pre-upgrade ritual from our setup guide, non-negotiable as always.
- Read the v3 release notes' breaking-changes section against your setup — the API and mobile-app version pairing matter; update server and apps together.
- Pull, recreate, and let database migrations finish before judging performance — first-start work on a big library is normal.
- Then do the fun part in order: one Workflow, JPEG XL for new uploads, a mobile curation session. In that order, over a week — not all in the first hour.
What to skip (for now)
Skip rebuilding your entire library organization around Workflows in week one — let a few rules prove themselves. Skip mass format experiments on originals (Immich never touches originals; keep it that way in your own scripts too). And skip the urge to self-host a second photo system 'for comparison' — the v3 line is where the category's center of gravity clearly sits, and your curation energy is better spent inside it.
Official documentation and references
- Immich blog — the v3.0.0 release notes that introduced Workflows
- Immich documentation — supported formats, including JPEG XL assets
- Immich documentation — partner sharing
- Immich documentation — tags
- Immich documentation — the official docs home
Frequently asked questions
Are my original photos converted to JPEG XL?
No. Immich never modifies originals. JPEG XL applies to generated thumbnails and previews, which is where the aggregate storage savings live across a large library.
What are Immich Workflows, in one sentence?
Built-in if-this-then-that automation for your photo library (currently a preview feature): asset events like an upload completing trigger filters and actions — filing into albums, archiving, tagging, visibility changes — configured in the UI instead of external scripts. There is no face-recognition trigger yet, and transcoding is a background-job setting, not a workflow action.
Should I upgrade from v2 now?
Yes, with the standard ritual: database backup first, release-notes breaking changes checked, server and mobile apps updated together. v3.4 has had the early-adopter period; the features are the point of the platform now.

