IliasHad/edit-mind v0.14.2

on GitHub

Overview

This release focuses on performance improvements across the entire video processing pipeline, a major architectural shift to Docker-first deployment, and several important bug fixes.

Breaking Changes

• Desktop app removed: We are fully transitioning to Docker. The Electron desktop app has been removed in favor of the web-based interface running via Docker Compose. We may have a separate Desktop client in the near future but in a separate repo
• node-llama-cpp dropped: Local LLM support via node-llama-cpp has been removed. AI backend options are now limited to Ollama and Gemini. It's much easier to Ollama option than node-llama-cpp

New Features

1. Benchmarks Page: Added a new page to share and debug video processing performance metrics, including per-stage frame analysis metrics. http://localhost:3745/app/benchmarks
2. GPU Support for Embeddings: The embedding service now supports GPU acceleration via CUDA.
3. PyAV for Frame Extraction: Switched to pyav from opencv for significantly faster video frame extraction and support. Also, adding support NVDEC to extract frames faster over NIVIDA GPUs.
4. Scene Thumbnail Generation in Python: Moved scene thumbnail creation to the Python script from the scene creation Node.js queue for better performance and consistency. Also, to make sure that the thumbnail represent exactly which frame has been used for frame analysis

Bug Fixes

1. Fixed GPU support for the transcription model. int8_float16 issues. #54
2. Refactored and optimized Docker images for faster builds.
3. Fixed an issue with vector main collection metadata updates.
4. Fixed a bug where searching scenes by text query returned incorrect results.
5. Fixed JPEG quality settings for generated thumbnails.
6. Improved error handling around the progress callback to prevent silent failures.
7. Improved the onboarding flow for new users to let the system download ML models for ML service before start processing the videos. #86
8. Removed the ffmpeg-installer npm package. FFmpeg is now sourced directly from the Docker image. (It was used to support Docker and Desktop application)
9. Fixed various TypeScript type issues.

Performance Improvements

• Significant plugin performance optimizations.
• Improved video frame extraction pipeline.
• Stage performance metrics now tracked during frame analysis.

Full Changelog: v0.14.1...v0.14.2