cd /news/machine-learning/immich-3-3-photo-and-video-managemen… · home › topics › machine-learning › article
[ARTICLE · art-148172] src=linuxiac.com ↗ pub= topic=machine-learning verified=true sentiment=↑ positive

Immich 3.3 Photo and Video Management Solution Adds People Sharing, Birthday Memories

Immich released version 3.3, adding shared people management, birthday memories, automatic stacking of edited photos, and machine learning optimizations, the open-source self-hosted photo and video management project announced. The release adds support for the multilingual PP-OCRv6 model family in tiny, small, and medium variants, extends Rockchip RKNPU support to all models including OCR, and brings TensorRT RTX acceleration to NVIDIA Ampere and newer GPUs. Administrators can enable the optimized models by setting MACHINE_LEARNING_MODEL_REVISION=v2 in the machine learning container, and existing thumbnails must be regenerated to benefit from the new resampling method.

by read2 min views1 publishedOct 9, 2026
Immich 3.3 Photo and Video Management Solution Adds People Sharing, Birthday Memories
Image: Linuxiac (auto-discovered)

Immich, the popular open-source, self-hosted photo and video management solution and privacy-focused alternative to Google Photos, has released version 3.3, introducing shared people management, birthday memories, automatic stacking of edited photos, and improvements to machine learning performance.

One of the main additions is the ability to share recognized people with other users in the same cluster group. Once shared, a person appears in both users’ people lists, including in the web interface and mobile apps. Both users can also edit the person’s name and birth date.

Sharing works in one direction, so users must share people back to enable two-way synchronization. Users can also choose whether changes to names and birth dates apply to everyone with shared access or only to their own account.

Another new feature is birthday memories. When someone in the photo library has a birthday, Immich can now automatically create a special memory featuring photos from their previous birthdays.

Photo organization improves with automatic stacking of mobile edits. When users edit pictures using external smartphone apps, Immich groups the edited versions with the originals instead of showing them as separate entries in the timeline.

Moreover, version 3.3 introduces a more accurate resampling method for thumbnails and previews, improving detail preservation and brightness, especially in high-contrast areas. The change applies to all supported image formats, but existing thumbnails must be regenerated to benefit.

Machine learning also receives attention. Developers optimized existing models to reduce memory consumption and improve processing speed across supported backends, including CPU-based inference. Support for Rockchip’s RKNPU now covers all models, including OCR, while NVIDIA Ampere and newer GPUs benefit from TensorRT RTX acceleration.

These optimizations are optional and can be enabled by setting MACHINE_LEARNING_MODEL_REVISION=v2 in the machine learning container. Since the optimized models produce equivalent results, you don’t need to rerun existing machine learning jobs.

Speaking of OCR, Immich 3.3 adds support for the multilingual PP-OCRv6 model family, available in tiny, small, and medium variants. The new models offer improved text recognition accuracy, with options suitable for both modest and more powerful hardware.

Administrators also get improved OAuth integration. User roles, storage quotas, and preferred usernames can now synchronize with an identity provider on every login, not just during initial registration.

Among the smaller enhancements, album editors can now update album titles and descriptions, JFIF image support has been added, and non-Latin text rendering on maps has been improved.

Finally, the release addresses numerous bugs affecting mobile backups, facial recognition, video transcoding, and the web interface. A long-standing issue with incorrect storage capacity reporting on macOS, particularly in some Docker configurations, has also been resolved.

For more information, see the release notes on GitHub. Image credits: Immich

── more in #machine-learning 4 stories · sorted by recency
── more on @immich 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/immich-3-3-photo-and…] indexed:0 read:2min 2026-10-09 · —