zrt-ai-lab/ViNote

What it solves

ViNote transforms videos from various platforms (YouTube, Bilibili) or local files into structured knowledge assets. It eliminates the manual effort of watching long videos for information extraction by automating the process of transcription, summarization, and organization.

How it works

The system uses a pipeline that combines video downloading (via yt-dlp) and audio transcription (via Faster-Whisper). It then leverages Large Language Models (LLMs) to process the transcribed text into structured notes, summaries, and mind maps. It also features a conversational agent, ViNoter, which allows users to perform these tasks—searching, transcribing, and summarizing—through natural language dialogue. For efficiency, it implements a "subtitle-first" strategy, prioritizing existing platform subtitles to skip the time-consuming ASR process when available.

Who it’s for

• Students and Researchers: For quickly extracting key points from educational videos and tutorials.
• Content Creators: For organizing video content into written formats or knowledge cards.
• Knowledge Workers: For building a personal knowledge base from video-based learning materials.

Highlights

• ViNoter Agent: A conversational interface for end-to-end video processing (search $\rightarrow$ transcribe $\rightarrow$ note $\rightarrow$ translate).
• Multi-format Output: Generates Markdown notes, interactive mind maps (via Markmap), and specialized knowledge cards.
• Intelligent Video QA: An AI-powered Q&A system that can answer questions based on specific video content or a collection of existing notes.
• Broad Compatibility: Supports YouTube, Bilibili, and local video files (MP4, AVI, MOV, MKV, etc.).
• Knowledge Management: Includes a tagging system, 17 preset categories, and SQLite persistence for organizing large volumes of notes.