Video Scribe

Video Scribe is a web application that processes video files, extracting audio, transcribing speech, and capturing key frames. It provides a visual interface for analyzing video content and generating markdown-based transcripts with embedded images.

Features

• Video upload and processing
• Audio extraction and transcription
• Key frame extraction
• Interactive transcript editing
• Markdown export with embedded images
• Dark mode support

Prerequisites

• Node.js (v14 or later)
• FFmpeg

Usage

No installation required. Start the server via:

npx videoscribe

or:

deno run --allow-read --allow-write --allow-net --allow-env --allow-run https://raw.githubusercontent.com/gramener/videoscribe/refs/heads/main/cli.js

Log into LLM Foundry.

Then open your browser and navigate to http://localhost:3000 and upload a video file.

This will transcribe the audio and extract key frames from the video. You can then:

1. View and edit the transcript
2. Play the extracted audio
3. Toggle key frames on/off
4. Export the result as a ZIP file with the keyframes, Markdown transcript and JSON transcript

API Endpoints

• POST /audio: Extract audio from uploaded video
• POST /keyframes: Extract key frames from uploaded video

Both endpoints accept multipart form data with a file field containing the video file.

Technologies Used

• Backend: Node.js, Express.js
• Frontend: HTML, CSS, JavaScript (ES6+)
• UI Framework: Bootstrap 5
• Templating: lit-html
• Audio Processing: FFmpeg
• Transcription: Groq API (distil-whisper-large-v3-en model)

License

This project is licensed under the MIT License.