This project provides an extension toolkit for Immich, enabling advanced photo management through AI-powered similarity detection and duplicate identification. It targets Immich users seeking to efficiently clean and organize their photo libraries by leveraging visual similarity analysis.

How It Works

The toolkit fetches user and asset data from the Immich PostgreSQL database, converts images into feature vectors using ResNet152, and stores these vectors in a Qdrant vector database. It then utilizes vector similarity to identify duplicate or visually similar photos, displaying them in a web-based UI for management. Deletions are performed by updating the Immich database directly, respecting Immich's trash feature.

Quick Start & Requirements

• Installation: Docker Compose (CPU/GPU) or Source Installation.
• Prerequisites: An Immich installation with the trash feature enabled. Database connection details for Immich. For GPU support: NVIDIA GPU with CUDA, NVIDIA Docker runtime, and Linux.
• Setup: Docker Compose offers a one-click install. Source installation requires manual setup of Qdrant and Python dependencies.
• Documentation: README

Highlighted Details

• AI-powered similarity detection using ResNet152 and Qdrant.
• Features include duplicate detection, batch operations, and filtering.
• Supports multi-mode search for processing multiple groups of similar photos simultaneously.
• Offers "Related Tree" mode for comprehensive similarity analysis within a single group.

Maintenance & Community

• Actively developed by RazgrizHsu. Contributions are welcome via Pull Requests.

Licensing & Compatibility

• Licensed under GNU General Public License v3.0 (GPLv3). Commercial use is permitted, but derivative works must also be open-sourced under GPLv3.
• Version compatibility between MediaKit and Immich is crucial due to database schema changes (singular vs. plural table names).

Limitations & Caveats

• Requires careful version matching between MediaKit and Immich to avoid database errors.
• The tool is under active development and may contain unexpected behaviors; database backups are recommended before use.
• AI-based similarity detection may not perfectly match human perception of similarity.