v3.3.0

[immich-push-o-matic[bot]]

v3.3.0

Note

We're collecting feedback related to people sharing in this discussion: #32196

Welcome to Immich v3.3.0!

This release includes people sharing, birthday memories, auto-stacked mobile edits, and a variety of other enhancements and bug fixes. Keep reading below for the full list of highlights.

Highlights

• Shared people management
• Birthday memories
• Auto-stacked mobile edits
• Sync OAuth claims on every login
• Higher quality thumbnails and previews
• Machine learning performance and accuracy improvements (opt-in)
• New OCR models
• Notable fix: correct storage size on MacOS

Shared people management

We are very pleased to release the next big milestone towards better sharing: people can now be shared!

As a user you can share people with another user in your cluster group:

Note

The “Manage access” modal can be opened from both the “People” (/people) and “Sharing settings” (/user-settings) pages in the web application. The modal is not available on mobile, yet.

When sharing a person with another user, the shared person:

• Shows up in both users’ list of people (on mobile and web)
• Shows up in the asset viewer / asset detail view on both owned and shared assets
• Can be edited by both users (only the name & birth date fields)

Note

Person sharing is directional. This means that each user needs to share each person back for edits to automatically work in both directions.

By default, changes to the name & date of birth are reflected for both users. On the web, the “Edit person” modal has an option to change the name/date of birth for “Only me” instead of the whole group.

Also, there is a preference for this behavior in the “Sharing settings”, which can be changed to make this the default behavior when editing a person. The setting can be changed here: https://my.immich.app/user-settings?isOpen=feature+people.

The setting has two options:

• “Everyone with shared access” — means an update will apply for everyone, covering the typical family use case where one or two users are supposed to be responsible for the entire family's people.
• “Only me” — means the names and dates of births will remain unchanged for other users

Lastly, by allowing cross-user merging, you can now merge a face from your own assets with people shared with you from other users. This is especially helpful if you didn't want to re-run facial recognition after joining a cluster group before. Now, after setting up bi-directional sharing, users can manually link people together.

Birthday memories

In an effort of making Immich more joyful to use, we are introducing memories for people's birthdays.

If it's the birthday of a person in your library, Immich will now celebrate that with a special memory featuring previous birthdays.

Auto-stacked mobile edits

When using something other than Immich for editing images on your phone, Immich will now stack those edits on top of the original. This allows for Immich to show all edits in one place, without cluttering the timeline.

This should provide a more organized timeline, especially when you heavily rely on editing assets on your phone.

Sync OAuth claims on every login

Immich supports OAuth claims for storage quota, the user role, and the preferred username. Up until now, those were only synced on registration.

Now, they can be synced on every login, allowing you to do active user management through your IDP, updating quotas and roles as you go and as necessary, and Immich will adopt those changes.

Higher quality thumbnails and previews

Immich now uses a more accurate image resampling method, improving detail preservation and brightness, especially for high contrast details. All image formats benefit from this change.

Before

After

Thumbnails already generated before this release won’t be affected by this change. To make it apply to existing thumbnails, you can run thumbnail generation on all assets in the job panel.

Machine learning performance and accuracy improvements (opt-in)

Note

We have re-exported and published the same machine learning models in an optimized format. This means they can be used as-is and do not require re-running any jobs. Opt-in to the revised machine learning models by adding the following environment variable to the machine learning container: MACHINE_LEARNING_MODEL_REVISION=v2.

After a long list of optimizations, ML is both significantly faster and uses less memory. This affects every backend, including CPU inference. However, the level of improvement varies by model and backend. Cases where certain backends (such as OpenVINO) produced wrong outputs at times should be resolved.

RKNPU, used by Rockchip boards, now supports every model in the catalog including OCR for the first time, bringing it to parity with other backends. Additionally, TensorRT RTX is now used for Ampere (30xx) and newer NVIDIA GPUs, improving performance significantly.

To fully benefit, set MACHINE_LEARNING_MODEL_REVISION=v2to allow more optimized models to be downloaded and used. Many of the improvements rely on these optimized models and will not take effect when using the old ones. The outputs of the new and old models are numerically equivalent, so switching doesn’t require re-running any tasks. This setting will be the default in a later release.

As always, keep in mind that the very first time you load a model requires more preparation work, so don’t be surprised if it takes some time at first. This preparation is cached and reused, so loading the model will be much faster after the first time.

New OCR models

The PP-OCRv6 family is now supported. These models are significantly more accurate than the previous PP-OCRv5 generation while all being multilingual. PP-OCRv6_tiny is efficient and suitable for modest hardware, PP-OCRv6_small is a notable jump in accuracy compared to PP-OCRv5_mobile, and PP-OCRv6_medium improves on PP-OCRv5_server as the highest quality and most demanding option.

Notable fix: correct storage size on MacOS

The block size used to calculate the storage capacity and usage has some nuances to it. In some situations (*cough* Docker on MacOS *cough*), we didn’t have enough information to correctly compute accurate values. With this release, the reported storage usage and capacity should now be correct.

Technical details

Recent NodeJS releases now expose the frsize property, which is sometimes different from bsize (“block size”). In those cases the calculate storage sizes could often be off by a factor of4096 (or more). In fact, #4318 has been our fourth oldest open issue (including the renovate dashboard), opened just over three years ago. Now, with accurate block sizes our storage calculation should be correct, even in Docker on MacOS.

We are very happy to finally see this issue resolved, even though it only meant a dependency bump and a tiny code change for us.

As always, please consider supporting the project.

🎉 Cheers! 🎉

Support Immich

If you find the project helpful, you can support Immich by purchasing a product key at https://buy.immich.app or our merchandise at https://immich.store

---

And as always, bugs are fixed, and many other improvements also come with this release.

What's Changed

🚀 Features

• feat(mobile): stack edited photos over the original on upload by @santoshakil in feat(mobile): stack edited photos over the original on upload #31082
• feat: birthday memories by @alextran1502 in feat: birthday memories #30831
• feat: people management by @danieldietzler in feat: people management #31620
• feat(ml): trt-rtx by @mertalev in feat(ml): trt-rtx #31863

🌟 Enhancements

• feat(web): improve non-Latin text rendering on the map by @meesfrensel in feat(web): improve non-Latin text rendering on the map #31611
• feat: improved storage template onboarding by @bwees in feat: improved storage template onboarding #31531
• feat(web): default the activity log to show the latest by @nekorevend in feat(web): default the activity log to show the latest #28938
• feat(server): linear-light resampling by @mertalev in feat(server): linear-light resampling #31236
• feat(web): add leave album action by @kojomba in feat(web): add leave album action #31768
• feat(ml): improve model loading by @mertalev in feat(ml): improve model loading #31742
• feat(ml): optimized model inference by @mertalev in feat(ml): optimized model inference #31750
• feat: editors can update album title & description by @jrasm91 in feat: editors can update album title & description #31805
• feat(server): add support for .jfif files by @Trithereon in feat(server): add support for .jfif files #31778
• feat: memory UI for birthday by @alextran1502 in feat: memory UI for birthday #31798
• fix(server): use GeoNames for reverse-geocoded country names by @rufusutt in fix(server): use GeoNames for reverse-geocoded country names #30199
• feat: extract manufacturer and device for samsung videos by @jonastahl in feat: extract manufacturer and device for samsung videos #29375
• feat(server): sync OAuth claims on login (OAuth immich_quota claim does not updates user quota on login #8073) by @thomasdelorge in feat(server): sync OAuth claims on login (#8073) #29013
• feat(mobile): editors can update album title & description by @shenlong-tanwen in feat(mobile): editors can update album title & description #31954
• feat(web): add shortcut to remove assets from an album by @cratoo in feat(web): add shortcut to remove assets from an album #31940
• feat(ml): ppocr v6 by @mertalev in feat(ml): ppocr v6 #32147

🐛 Bug fixes

• fix: Build SDK in dev container by @Pecacheu in fix: Build SDK in dev container #31096
• fix(ml): race when submitting to rknn execution queue by @swbchangle in fix(ml): race when submitting to rknn execution queue #31143
• fix(mobile): use timeline scroll velocity to add placeholders by @agg23 in fix(mobile): use timeline scroll velocity to add placeholders #29443
• fix(ml): openvino config has unnecessary devices and volumes by @mertalev in fix(ml): openvino config has unnecessary devices and volumes #31253
• fix(mobile): rewrite slideshow controller system by @agg23 in fix(mobile): rewrite slideshow controller system #30771
• fix: make local album provider reactive by @shenlong-tanwen in fix: make local album provider reactive #31582
• fix(web): trim strings before searching/comparing by @meesfrensel in fix(web): trim strings before searching/comparing #31614
• fix(mobile): load more than one image on Android with Remove animations by @agg23 in fix(mobile): load more than one image on Android with Remove animations #31587
• fix(mobile): make EXIF provider reactive by @agg23 in fix(mobile): make EXIF provider reactive #31552
• fix: set limit maximum zoom in asset viewer by @TokenLimitExceeded in fix: set limit maximum zoom in asset viewer #31609
• feat: asset face v3 by @jrasm91 in feat: asset face v3 #31591
• fix(mobile): fill video placeholder to prevent flicker on open by @LeLunZ in fix(mobile): fill video placeholder to prevent flicker on open #31621
• fix(mobile): show the real date for android local photos with no exif by @santoshakil in fix(mobile): show the real date for android local photos with no exif #29193
• fix(mobile): prevent live photo from getting stuck during dismiss animation by @LeLunZ in fix(mobile): prevent live photo from getting stuck during dismiss animation #28080
• perf: faster server cold start by @holoskii in perf: faster server cold start #31630
• fix(web): correct memory title date by @Dhruthiya in fix(web): correct memory title date #31643
• fix(mobile): clamp out of range datetimes so timeline queries cannot crash by @santoshakil in fix(mobile): clamp out of range datetimes so timeline queries cannot crash #30490
• fix(web): scrollable search dropdown on mobile web by @meesfrensel in fix(web): scrollable search dropdown on mobile web #31634
• fix(web): queue buttons focus outline by @meesfrensel in fix(web): queue buttons focus outline #31657
• fix(mobile): ios clear file cache feedback by @yranandika05 in fix(mobile): ios clear file cache feedback #31401
• fix(mobile): prevent SurfaceLayer crashing on maps in Viewer on Android by @agg23 in fix(mobile): prevent SurfaceLayer crashing on maps in Viewer on Android #31709
• fix: preserve scroll position on memories page by @YarosMallorca in fix: preserve scroll position on memories page #31588
• fix(server): compare integrity paths independently of unicode normalization by @justadityaraj in fix(server): compare integrity paths independently of unicode normalization #31102
• fix(mobile): retry the background backup when connectivity is restored by @santoshakil in fix(mobile): retry the background backup when connectivity is restored #30909
• fix(web): refresh people after setting person thumbnail by @meesfrensel in fix(web): refresh people after setting person thumbnail #31744
• fix: save user_version in migration transaction by @shenlong-tanwen in fix: save user_version in migration transaction #31039
• fix(mobile): play the server copy when the local video file cannot be read by @santoshakil in fix(mobile): play the server copy when the local video file cannot be read #31606
• fix(mobile): resume after the app was launched in the background by @santoshakil in fix(mobile): resume after the app was launched in the background #31557
• fix(web): prevent race when editing album description during upload by @dikshit-n in fix(web): prevent race when editing album description during upload #31758
• fix(server): attach error handler to accepted sockets to prevent crash on connection reset by @aashish00021 in fix(server): attach error handler to accepted sockets to prevent crash on connection reset #31079
• fix(mobile): ask before deleting local only photos on android 10 and below by @santoshakil in fix(mobile): ask before deleting local only photos on android 10 and below #31146
• fix(web): remember selected search type by @DawidKrynski in fix(web): remember selected search type #31799
• fix: face bounding box label alignment by @danieldietzler in fix: face bounding box label alignment #31793
• fix(server): incorrect scaling of rotated videos when transcoding by @fabxyz in fix(server): incorrect scaling of rotated videos when transcoding #31473
• fix(web): keep other comments visible when deleting a comment by @DawidKrynski in fix(web): keep other comments visible when deleting a comment #31848
• fix(server): write tag updates to exif table by @jorbrock in fix(server): write tag updates to exif table #31425
• fix(server): exclude hidden assets from large files search by @bo0tzz in fix(server): exclude hidden assets from large files search #31479
• fix: favorite workflow step description by @xCJPECKOVERx in fix: favorite workflow step description #31840
• fix(ml): stricter typing and request validation by @mertalev in fix(ml): stricter typing and request validation #31856
• fix(server): emit sidecarwrites and assetupdates consequently by @meesfrensel in fix(server): emit sidecarwrites and assetupdates consequently #31777
• fix: library watcher should respect ignore patterns by @etnoy in fix: library watcher should respect ignore patterns #31418
• fix: email template by @danieldietzler in fix: email template #31859
• fix(web): cull offscreen thumbnails on individual shared links by @bo0tzz in fix(web): cull offscreen thumbnails on individual shared links #31890
• fix: person management ids validation and permissions by @danieldietzler in fix: person management ids validation and permissions #31898
• fix(server): person birthdate timezone bug by @jrasm91 in fix(server): person birthdate timezone bug #31901
• fix(mobile): face thumbnail stuck blank after renaming a person by @santoshakil in fix(mobile): face thumbnail stuck blank after renaming a person #30769
• fix(server): make "limit" paramter use z.coerce by @aviv926 in fix(server): make "limit" paramter use z.coerce #31899
• fix(server): dont queue unneeded jobs when no xmp file exists by @etnoy in fix(server): dont queue unneeded jobs when no xmp file exists #31927
• fix(server): ignore Samsung Gallery rotation tag for HEIF orientation by @bo0tzz in fix(server): ignore Samsung Gallery rotation tag for HEIF orientation #31937
• fix: restore folder view scroll position by @raahu1l in fix: restore folder view scroll position #31860
• fix(ml): widen empty half-precision arrays by @mertalev in fix(ml): widen empty half-precision arrays #31950
• fix(ml): serialize trt-rtx sessions by @mertalev in fix(ml): serialize trt-rtx sessions #31951
• fix(server): make library exclusion matching consistently case-insensitive by @etnoy in fix(server): make library exclusion matching consistently case-insensitive #31970
• fix: stale memory thumbnail by @danieldietzler in fix: stale memory thumbnail #31975
• fix: storage usage display on virtiofs and alike by @danieldietzler in fix: storage usage display on virtiofs and alike #31982
• fix(server): include album owner for album shared links by @meesfrensel in fix(server): include album owner for album shared links #31837
• fix(mobile): don't re-upload a photo the server already has by @santoshakil in fix(mobile): don't re-upload a photo the server already has #31941
• fix(mobile): make the iOS background batch wait for its remote sync by @santoshakil in fix(mobile): make the iOS background batch wait for its remote sync #31942
• fix(mobile): don't start the foreground backup in iOS background launches by @santoshakil in fix(mobile): don't start the foreground backup in iOS background launches #31943
• fix(mobile): image jumps when full-res image loads during pinch or zoom animation by @LeLunZ in fix(mobile): image jumps when full-res image loads during pinch or zoom animation #31948
• fix(ml): upgrade to openvino 2026.4.1 by @mertalev in fix(ml): upgrade to openvino 2026.4.1 #31996
• fix(cli): sidecar precedence by @etnoy in fix(cli): sidecar precedence #31929
• fix: metadata extraction for media streams with unknown profile by @danieldietzler in fix: metadata extraction for media streams with unknown profile #32007
• fix: assetTagFilter any matching by @danieldietzler in fix: assetTagFilter any matching #32008
• fix: improper filename sanitizing by @danieldietzler in fix: improper filename sanitizing #32009
• fix(server): don't select an audio stream ffprobe could not identify by @RxChi1d in fix(server): don't select an audio stream ffprobe could not identify #30901
• fix: stale memory cover by @danieldietzler in fix: stale memory cover #32098
• fix(web): no trash actions when trash is empty by @meesfrensel in fix(web): no trash actions when trash is empty #32097
• fix: bounding box label clipping due to reactivity issue by @danieldietzler in fix: bounding box label clipping due to reactivity issue #32090
• fix: people page search by @danieldietzler in fix: people page search #32094
• fix(web): album reactivity by @meesfrensel in fix(web): album reactivity #32095
• fix(ml): incorrect outputs on alchemist and battlemage by @mertalev in fix(ml): incorrect outputs on alchemist and battlemage #32110
• fix: memory upcoming query by @jrasm91 in fix: memory upcoming query #32164
• fix(web): show face name label above box when it would be hidden by @joepadmiraal in fix(web): show face name label above box when it would be hidden #32088

📚 Documentation

• chore(docs): deprecate outlook SMTP by @mmomjian in chore(docs): deprecate outlook SMTP #31139
• feat: new FAQ entries by @bo0tzz in feat: new FAQ entries #31227
• fix(docs): add needed mobile app developer setup step by @arth3mis in fix(docs): add needed mobile app developer setup step #31263
• docs(server): document the UTC timestamp format for timeBucket by @alangrainger in docs(server): document the UTC timestamp format for timeBucket #31421
• chore(docs): add note about opening issues to contributing by @meesfrensel in chore(docs): add note about opening issues to contributing #31607
• docs: link to the docker compose builder by @bo0tzz in docs: link to the docker compose builder #31610
• fix: page path in better face clusters guide by @bo0tzz in fix: page path in better face clusters guide #31671
• docs: document locked folder session behaviour by @howroyd in docs: document locked folder session behaviour #31735
• docs: update backup settings menu path to Database Dump Settings by @MalibThalion in docs: update backup settings menu path to Database Dump Settings #31270
• docs: add QNAP install guide by @hraabis in docs: add QNAP install guide #30695

🌐 Translations

• chore(web): update translations by @weblate in chore(web): update translations #31376
• chore(web): update translations by @weblate in chore(web): update translations #31864

New Contributors

• @Pecacheu made their first contribution in fix: Build SDK in dev container #31096
• @swbchangle made their first contribution in fix(ml): race when submitting to rknn execution queue #31143
• @maximal made their first contribution in chore(web): Change x to typographically correct × (× in HTML) symbol #31304
• @TokenLimitExceeded made their first contribution in fix: set limit maximum zoom in asset viewer #31609
• @holoskii made their first contribution in perf: faster server cold start #31630
• @lg114 made their first contribution in fix(docs): update README_zh_CN: typos, links, cleanup  #31654
• @Dhruthiya made their first contribution in fix(web): correct memory title date #31643
• @howroyd made their first contribution in docs: document locked folder session behaviour #31735
• @dikshit-n made their first contribution in fix(web): prevent race when editing album description during upload #31758
• @aashish00021 made their first contribution in fix(server): attach error handler to accepted sockets to prevent crash on connection reset #31079
• @nekorevend made their first contribution in feat(web): default the activity log to show the latest #28938
• @kojomba made their first contribution in feat(web): add leave album action #31768
• @DawidKrynski made their first contribution in fix(web): remember selected search type #31799
• @Trithereon made their first contribution in feat(server): add support for .jfif files #31778
• @fabxyz made their first contribution in fix(server): incorrect scaling of rotated videos when transcoding #31473
• @rufusutt made their first contribution in fix(server): use GeoNames for reverse-geocoded country names #30199
• @avaly made their first contribution in chore: log assets deletion in background job #31770
• @MalibThalion made their first contribution in docs: update backup settings menu path to Database Dump Settings #31270
• @jonastahl made their first contribution in feat: extract manufacturer and device for samsung videos #29375
• @thomasdelorge made their first contribution in feat(server): sync OAuth claims on login (#8073) #29013
• @raahu1l made their first contribution in fix: restore folder view scroll position #31860
• @hraabis made their first contribution in docs: add QNAP install guide #30695
• @ChocolateChipKookie made their first contribution in fix(web): don't swallow Escape when person name editor is closed #32123
• @joepadmiraal made their first contribution in fix(web): show face name label above box when it would be hidden #32088

Full Changelog: v3.2.4...v3.3.0

---

This discussion was created from the release v3.3.0.

[maybeanerd]

First off, great work on this release, a lot of cool new things!!

Now, they can be synced on every login, allowing you to do active user management through your IDP, updating quotas and roles as you go and as necessary, and Immich will adopt those changes.

Is there a way to disable e.g. the storage quota always being used? This release note reads like it could be optional ("they can be synced") but I didnt find any config settings for it in the oauth docs.

My usecase is that I want a default size (10GB) for all users, but then allow more storage in the immich app itself.
Currently im handling that by not configuring any quota in authentik, and just having a fallback in the oauth settings to 10GB

Im understanding the change now so that whenever users log in again, it will be reset to 10GB for all of them?

[jrasm91]

That should probably say "are synced on every login" since can't be turned off.

Previously if the identity provider changed a user's quota from 5GiB to 10GiB nothing would be updated in Immich. It ended up only being a "quota on user create" claim. Now, it updates on every login. If no quota is provided in the IdP or if the claim stopped being sent, nothing happens on the Immich side. We have discussed updating the quota to the default quota from the oauth settings, which currently only acts as a "quota on user create" mechanism.

[maybeanerd]

which currently only acts as a "quota on user create" mechanism.

Oh but this is nice, that's exactly what I'm looking for! I think this is a good way to have a default quote for newcomers while still allowing potential overrides through the quota from the IdP itself

[wardwygaerts]

This is especially helpful if you didn't want to re-run facial recognition after joining a cluster group before.

Does this mean we can add users to the cluster group without rerunning the facial recognition? This was resuired in the previous release, correct?

[jrasm91]

Yes that correct. Re-running facial recognition is not required. More information available in #32196

[AdrianK258]

Anyone else missing button for that shared Manage access for people? . I don't have that access in web interface.

[CarrotManMatt]

Same issue here that I don't have the button

[jrasm91]

Ok I just checked the code. It does only show up when you have at least one other user in your cluster group. Cluster groups can be managed in user settings > sharing.

[AdrianK258]

Ok I just checked the code. It does only show up when you have at least one other user in your cluster group. Cluster groups can be managed in user settings > sharing.

I have. My wife and myself. Yet its still not there

[danieldietzler]

If you invited your wife, did your wife accept the invite on her account? @AdrianK258

[AdrianK258]

If you invited your wife, did your wife accept the invite on her account? @AdrianK258

That was the issue. She is using app. So i had to tell her to log in thru web. Thank you.

I was thinking its sufficient if admin do it. 😁

[matticas]

PPPPPEEEEEEOOOOOOOOOPPPPPPPPPPPLLLLLLLLLLEEEEEEEEEEEEEEEEE SSSSSSSSSSSHHHHHHHHAAAAAAAAAAARRRRRRRIIIIIIIINNNNNNNNNGGGGGGGG!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

I've been waiting for this for so long!!!!!!!!! AMAZING!!!

THANK YOU SO MUCH!

Side note, I noticed my existing v3.2.4 I could view 'new' people which were my wife's account people. I merged them for fun and it worked on web, but not mobile.

Updating now and exploring the new features!

I've been looking forward to this for so long.

[jrasm91]

Yeah, there were some bugs with the mobile sync for this new functionality, but it should be fixed now. Let us know if you run into any issues with anything people sharing related in this discussion: #32196

[Str1atum]

Does the Birthday feature have to be enabled somewhere? I have a person with the birthdate of today and container date also matches, but no birthday card appears anywhere.

[danieldietzler]

It does not. Memories are generated a few days in advance though. That is, the memory for today would've already been generated 3(?) days ago, thus before v3.3 brought birthday memories. The next birthday should show up though :D @Str1atum

[yann117]

@Str1atum If you cannot wait or for testing purpose, you should be able to force the memories generation immediately as follow:

• in the Administration console
• on the left pane, select Job Queues
• on the right, select Create job
• in the list, select Memory generation
• wait for its execution, then refresh

[danieldietzler]

@yann117 that is incorrect. That won't generate memories for today either

[jrasm91]

We (plan to) add a button to allow (forcefully) regenerating a memory for a specific day on the /memories page. That would have solved this problem.

[iheartpizza1]

Wow what an amazing milestone! Thank you guys!

I was one of the weirdos with Docker on MacOS, having to explain to the family why they can ignore the storage usage widget 😅 Thank you for addressing this!
Unfortunately, it's still not accurate. The widget shows +30GB used compared to what df -h reports.


Is there a plan to allow the storage widget to display usage from the Immich data dir when it's on NFS? That's my case. It could be nice to see the actual data dir usage (the NFS mount) and not the docker host storage usage.

Again thank you and keep up the amazing work! 🚀 🥳

[danieldietzler]

It will stat the library folder. Wherever that is will be reported in this widget. So if you just make sure that folder is mounted to your NFS share, it should report the correct storage.
As for the 30GiB difference, that's really strange. Especially, as the available storage is spot on. Would probably need some more digging. If you feel like it, you could jump on our Discord and we can check it out :) It's also possible this won't be a problem for the NFS mount though

[antlarr]

Great release. Thanks for all the work into people sharing!

I've started using it and so far it works great (without rerunning the facial recognition), but I wonder... let's say there are two immich users (A and B) who have one person (Foo) tagged and shared mutually (A shares Foo with B and B shares their version of Foo with A), then they merge the faces (since it's the same person). Is it allowed for one of the immich users to stop sharing Foo with the other user? What would happen if they do it?

Thanks again!

[danieldietzler]

Yes user B may stop sharing. Naturally, the user A cannot edit user B anymore then. But, as long as they are still in the same cluster group, user A is going to see their own people (including the one that had been temporarily shared) in shared assets still. So basically, nothing significant is going to change.

[alexwhitman]

Is there a way to re-generate all thumbnails of existing images so that the new resampling method is used?

[danieldietzler]

Yes, re-run all for the thumbnail generation jobs in the admin settings https://my.immich.app/admin/queues @alexwhitman

[alexwhitman]

Excellent, thanks!

[yann117]

Regarding the opt-in Machine learning new model:

Machine learning performance and accuracy improvements (opt-in)
Note
Opt-in to the new machine learning models by adding the following environment variable to the machine learning container: MACHINE_LEARNING_MODEL_REVISION=v2

This is what I added in my compose.yml for the immich-machine-learning service:

  immich-machine-learning:
    # temporary change for the opt-in ML v2
    environment:
      MACHINE_LEARNING_MODEL_REVISION: 'v2'

Is that ok, how to verify it is correctly using the new model? I see in the log some reference to openai/v2:

[10/08/26 10:54:54] INFO     Downloading textual model 'ViT-B-32__openai' to    
                             /cache/clip/ViT-B-32__openai/v2/textual. This may  
                             take a while.                                      
Fetching 13 files:   8%|▊         | 1/13 [00:00<00:01,  6.42it/s]Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
[10/08/26 10:54:54] WARNING  Warning: You are sending unauthenticated requests  
                             to the HF Hub. Please set a HF_TOKEN to enable     
                             higher rate limits and faster downloads.           
Fetching 13 files: 100%|██████████| 13/13 [00:06<00:00,  2.15it/s]

As expected, the first search took some time, and then it was back to immediate/fast results.

[danieldietzler]

Yes, that's fine. Btw you may want to look into a better CLIP model. ViT-B-32__openai really isn't any good in comparison anymore. We have tables for different languages https://docs.immich.app/features/searching#clip-models

[yann117]

Waow, thank you for the hint. I guess that was the default from my very old setup long back ago; I never thought to change it.

I have now selected a multilingual model, which make the recall jump from 38.27 (ViT-B-32__openai) to 84.97 (ViT-L-16-SigLIP2-256__webli) 😄

[wardwygaerts]

Adding the value to my .env file is also fine I suppose?

[danieldietzler]

Did you mean to reply under this message? Which value? @wardwygaerts

[timglong]

Adding the value to my .env file is also fine I suppose?

That's how I did it, assuming you're referring to MACHINE_LEARNING_MODEL_REVISION=v2.

[eerison]

I don't know if it was introduced in v3.3.0.
But when I add a Partner, I can't remove anymore

The user below that appears "X" was already there.
The user above, I added in 3.3.0 , and "x" doesn't appear.

[danieldietzler]

Are you sure the user at the top isn't sharing with you?

[eerison]

Yep the user isn't sharing with me.

When the user share with me, appear a toggle button to show in my timeline.

But I checked and the user isn't sharing with me
Tested again in 3.3.1

[timglong]

Machine learning performance and accuracy improvements (opt-in)

I tested these with the last RC yesterday and the improvement is incredible. I re-ran all the ML jobs at the same time with my normal settings (ViT-SO400M-16-SigLIP2-256__webli, buffalo_l, PP_OCRv5_mobile) and the GPU (8GB RTX 5050) maxed out around 30% utilization.

30% util leaves quite a bit of headroom so I bumped the models to ViT-SO400M-16-SigLIP2-384__webli, antelopev2, and PP_OCRv5_server to see what would happen. Utilization jumped a bit but never went over 60%. A picture is worth a thousand words, so see below for the memory utilization of these models all running at the same time (bottom process) compared to the preload value for my remote ML instance (top process, still on -256__webli). Total time to process everything was ~23 minutes for ~11k assets.

Awesome work @mertalev and team!

ETA: I didn't even notice the OCRv6 part, nice!

[hh-pirat]

Title:
[BUG] Broken images, missing locations, and "0B of 0B" storage space after updating to v3.3.0 (Privileged Proxmox LXC)

Description:
Hi Immich Team,
I recently updated my Immich instance to version 3.3.0, and unfortunately, several critical parts of my installation broke immediately after the update.
The Bug / Symptoms:

1. Broken Images: All images that were uploaded or are in the library are no longer being displayed properly (showing placeholders/errors).
2. Missing Metadata/Locations: None of the geographical locations or metadata are being imported or assigned to the assets anymore.
3. Storage Bug: In the server/admin dashboard, the "Storage Space" indicator is completely broken, displaying exactly "0B of 0B" (0 bytes of 0 bytes).
   Please see the attached screenshots below illustrating the UI errors, the broken timeline, and the storage readout.
   (Note: You can drop your screenshots right here in the GitHub comment block)
   System Environment:
   • Immich Server Version: v3.3.0
   • OS / Deployment: Running natively inside a Privileged Proxmox LXC Container (Not using Docker).
   • Storage Setup: The storage paths / external libraries are mounted directly into the LXC.
   Additional Context / What I've noticed:
   Since this is a privileged Proxmox LXC, it seems like the database migration or a change in the microservices during the v3.3.0 update might have caused a breaking issue regarding storage paths, mount point parsing, or permissions, which causes the app to read 0B total storage and fail to locate file system assets or write cache data.
   Could you please advise on how to debug this or if there is a known issue with the storage-space parsing and systemd services on LXC instances for this release?
   Thank you for your hard work and help!

[danieldietzler]

@hh-pirat you are running with an unsupported deployment method

[antlarr]

I have another question about the people sharing feature. Another immich user shared a person with me and if I open that face and go to the "Manage Access" dialog, I see a button to "share back" that person... What does it do? I don't understand why would I share back a person with the user who actually owns it and shared it with me. (This is on 3.3.1, btw, but I think I also saw the button in 3.3.0). Thanks!