REST API, MCP server, token management
Release history
edit-mind releases
Local-first AI video intelligence platform. Index your video library with multi-modal analysis (YOLO, DeepFace, Whisper), search semantically via natural language, Docker-ready.
All releases
14 shown
- Upgraded project dependencies with latest security patches
- Custom metadata for face recognition
- Face attributes indexed in vector DB for rich searching
Full changelog
Overview
This release introduces custom face metadata enrichment to enable smarter video search, alongside security updates and updates to the README file, including a new Desktop App section and Edit Mind's featuring on Syntax.fm.
Features
Custom Metadata for Face Recognition
You can now attach rich, structured metadata to known faces and have it automatically indexed alongside face recognition results, a community-requested feature.
Place a metadata.json file inside each named face folder under your FACES_DIR, check the FACES_DIR environment variable in .env.system:
FACES_DIR/
└── Ilias/
├── photo.jpg
└── metadata.json
Example FACES_DIR/Ilias/metadata.json:
{
"name": "Ilias Haddad",
"alternate_names": ["Ilias Hadd"],
"nationalities": ["Netherlands", "Morocco"],
"date_of_birth": "1989-03-01",
"place_of_birth": "Dordrecht, Netherlands",
"height_m": 1.87,
"positions": ["Centre-back"],
"current_club": "Union Touarga",
"club_jersey_number": 20,
"dual_nationality": true
}
When a video is indexed and the face of "Ilias" is recognized in a scene, all metadata properties are embedded into the vector DB document for that scene. This means you can now search scenes using any metadata attribute — for example, searching current_club: Union Touarga will return every scene featuring that person.
Other Improvements
- Security updates: Upgraded project dependencies with the latest security patches.
- README updates: Added a Desktop App section and highlighted the Edit Mind feature on Syntax.fm.
Edit Mind featured at Syntax.fm
Watch Edit Mind featured on Syntax.fm (starts at 41:21)
Full Changelog: https://github.com/IliasHad/edit-mind/compare/v0.20.4...v0.21.0
- H.265 video transcoding support
- Background job FFmpeg route
- Folder job refresh on scan
Fixed MXF to MP4 transcoding by adding moov atom handling and improved file watcher to prevent incomplete video processing.
- NVDEC GPU codec support
- Auto-detection with CPU fallback
- Pagination UI for folders and jobs
Fixed embedding model file corruption handling by wrapping deletion logic for invalid protobuf files.
- MXF support requires docker-compose.yml media volume update with write permissions
- MXF video file support
- Automatic MP4 transcoding
- Improved onboarding with video downloads
- MAX_SCAN_DEPTH configuration
- ENABLE_QUEUE_UI for Bull Dashboard
- Custom video tagging with text embedding search
- DISABLE_AUDIO_EMBEDDINGS environment variable
- DISABLE_VISUAL_EMBEDDINGS environment variable
Fixed CUDA availability detection in web service and rescan button state management for Docker readiness.
- Desktop application removed in favor of Docker
- node-llama-cpp local LLM support removed
- GPU embeddings via CUDA
- PyAV for frame extraction
- Scene thumbnail generation in Python
- GPU-enabled Docker image with CUDA
- FFmpeg GPU acceleration
- Concurrent job limit configuration
- ML models caching for Ultralytics, Hugging Face, PyTorch
- Face recognizer accuracy improvements
- Immich face importer UI enhancement
- Multi-stage video processing pipeline
- Projects and Collections organization
- Image-based video search