I built an AI tool to summarize videos, useful for me, but would you use it? (github.com)

A developer released an open-source video-analysis pipeline that extracts transcriptions, representative-frame visual descriptions, and multimodal summaries block-by-block, then aggregates them into a final overview. It supports two modes: fully local (faster-whisper for STT, BLIP for vision, Ollama-hosted LLMs) for privacy/no-API-key use, or API-driven (Groq for STT/LLM + Google Gemini Vision for image descriptions) for cloud-powered models. The tool is designed for long clips using configurable BLOCK_DURATION (default 30s), language, summary size/persona, and extra prompts; it accepts local files or URLs (downloads via yt-dlp) and currently prints results to terminal but can be extended to JSON/SRT/Markdown or a web UI. Technically, the repo targets Python 3.10+, requires FFmpeg (audio extraction), OpenCV/Pillow, and offers optional GPU acceleration for Whisper. Core components include faster-whisper/WhisperModel, BLIP, Ollama (local LLMs), Groq and Google GenAI SDKs for API mode, and utilities for caching/resuming per-block results. Significance: it provides a practical, configurable multimodal summarization baseline for researchers and engineers working with long videos—balancing privacy and performance trade-offs, enabling modular model swaps, and encouraging reproducible experiments and extensions (model selection, multi-frame sampling, exports, CLI/UI).