Immich releases v3.4.0-rc.2 with machine learning and mobile bug fixes

Immich has published v3.4.0-rc.2, the second release candidate for its upcoming major version 3.4. This specific update is a maintenance release containing exclusively bug fixes for the server, web, and mobile components.

Users can opt into this release candidate by updating their Docker environment variables or using Testflight for iOS and beta tracks for Android. The release also highlights a transition in the machine learning subsystem to improve resource efficiency.

Technical improvements and bug fixes in v3.4.0-rc.2

• Machine Learning Optimization: Users can set the MACHINE_LEARNING_MODEL_REVISION environment variable to v2 to utilize improved memory usage and performance characteristics.
• Media Format Support: The iOS mobile application received fixes to correctly display 10-bit HEIC and AVIF original files.
• Server-side Logic: Improvements were made to OCR search text persistence during thumbnail regeneration and the validation of motion photo video ranges during extraction.
• Privacy and Telemetry: The update includes a fix to disable ONNX Runtime telemetry within the machine learning subsystem.
• Web and Mobile Stability: Fixes address OAuth regressions, web worker imports, websocket connectivity on the people page, and local database reset procedures on mobile.

Why it matters

As a popular self-hosted photo management solution, these refinements improve the reliability of automated media processing and machine learning tasks. The introduction of a new ML model revision allows sysadmins to better manage hardware resources when running Immich in containerized environments.