github open-noodle/gallery v5.11.1
v5.11.1 (immich v3.4.2)

46 minutes ago

Based on Immich v3.4.2

Release notes: https://opennoodle.de/releases/v5-11-0/

What's in this release

Faces across the whole video, and every moment a person appears

Face detection can now sample frames across a whole video instead of only its poster frame, and keeps the sharpest view of each person. Each face found in a video lists every moment that person appears, and each moment opens the video at that point; the moments show on the web and in the app, and on the web the owner can remove a moment or move it to another person. It's off by default: turn on Detect faces across the whole video in Administration → Settings → Machine Learning → Facial Recognition, then run Face Detection with Missing under Jobs. Named faces keep their person. (docs)

Faces found across a whole video

A person's moments in a video

Pets across the whole video

Pet detection can sample a whole video the same way, so a dog or cat that walks in late is found, shown from its clearest frame, with its moments listed like a person's. Turn on Detect pets across the whole video under Pet Detection, then run Pet Detection with Missing. Pets you have named keep their names. (docs)

Pet detection

The storage template on S3

On a server that keeps its photos in S3, the storage template now names and moves files in the bucket as it does on disk. An existing S3 library waits for one manual Storage Template Migration run.

Storage template on S3

Data Saver video on mobile data

The app can stream a small 720p copy of each video on mobile data. An administrator turns it on, and a switch in the app's video settings lets you opt out.

Data Saver video

Photos: Both, Personal or Spaces

Photos can show everything, only your personal photos, or only what reaches you through your Spaces, and a badge on each thumbnail shows which Spaces hold it. You can also remove photos from a Space right from the Photos timeline.

A bottom dock in the app

The app's tabs and the Years / Months / All switch now share one dock at the bottom of the screen. Scrolling down folds it into a single row, scrolling up brings it back, and it holds the display options too.

Also new

  • Switch a person to a pet, and back, when recognition files a cluster in the wrong list.
  • Stacks collapse in albums, as they do on the timeline.
  • Tags in the app and JPEG XL, from the Immich v3.4.2 base, with its security fixes.
  • Fixes: one person no longer splits into many people in a video, blurry video faces stay out of recognition, two parts of one person in a video merge again, Camera RAW files on S3 decode again, very thin images no longer break face detection, replaced motion videos are removed from S3, uploads holding only Exiv2 metadata or failing their checksum are turned away, map clusters list every photo, moving through a memory in the app is snappier, and the web face picker is ready to type into.

Android: install the APK

If Google Play hasn't offered you 5.11.1 yet, install the app from the attached gallery-v5.11.1.apk. The first time, Android will ask you to allow installs from your browser or files app.

This APK is built from the same commit as the server. It brings a person's moments in a video and the new bottom dock to the app.

Update

Back up your database first. Then set IMMICH_VERSION=v5.11.1 in your .env and run:

docker compose pull && docker compose up -d

AMD GPU users: machine learning is unchanged in this release, so keep the ROCm image ghcr.io/open-noodle/gallery-ml:v5.11.0-rocm.

To go back to an earlier Gallery release, follow Rolling back to an earlier release.

Report issues on GitHub.

Gallery changes

  • fix(web): focus the person picker's search field on open, and let Escape close it (#1289)
  • fix(server): reject Exiv2 stub uploads and checksum mismatches (#1294)
  • fix(web): keep the markers on a map cluster's edges in its panel (#1314)
  • fix(faces): give each sharper-view detection to one video face only (#1325)
  • feat(mobile): a bottom dock that folds on scroll (#1322)
  • fix(mobile): snappier taps and swipes between photos in a memory (#1318)
Other Gallery changes
  • ci(rc): build RC images for labelled PRs from forks (#1326)
  • ci(release): optionally reuse an earlier tag's OpenVINO ML image (#1324)
Upstream commits

Contributors

Thank you to everyone who contributed to this release:

Don't miss a new gallery release

NewReleases is sending notifications on new releases.