Video Note Taking

Elegant, modular tool to generate notes directly from videos, subtitles, or plain transcripts.

Quick start

Global Installation

• Install globally with pnpm:

pnpm i -g video-note-taking

• Ensure your Google Gemini API key is set:

export GEMINI_API_KEY=your_key

• Run:

vnt input.mp4 -o note.md

Alternative: Local Development

• Install dependencies and build:

pnpm install
pnpm build

• Run:

node dist/cli.js input.mp4 -o note.md

Usage

vnt <inputs...> -o note.md [options]

Options:
    -o, --output <file>     Output file path, or - for stdout (default: "note-<timestamp>.md")
    --style <style>         Note style (choices: "handwritten", "bulleted", "outline", "concise", default: "handwritten")
    -l, --language <lang>   Output language, e.g., zh-TW or en
    --timestamps            Include timestamps when available
    --merge                 Merge multiple inputs into one note
    --provider <name>       Provider name (default: "google-genai")
    --api-key <key>         Provider API key (or use env GEMINI_API_KEY)
    --model <name>          Provider model name
    --prompt <text>         Custom system instruction override
    --format <fmt>          Output format (md|txt) (default: "md")
    -h, --help              Show help

Library usage

You can programmatically use the core API:

import { generateNotes } from "video-note-taking";

const { content } = await generateNotes(["input.mp4"], {
  style: "handwritten",
  language: "English",
  providerOptions: { apiKey: process.env.GEMINI_API_KEY },
});

Notes

• Provider currently supports Google Gemini via @google/genai.
• Supported video MIME types include: mp4, mov, avi, mpeg/mpg, flv, webm, wmv, 3gpp.
• You can pass large video files directly (up to 2 GB per file via the Gemini Files API). The CLI uploads videos and asks Gemini to understand content and produce notes. Files are stored for ~48 hours by the API.
• If you also pass .srt or .txt, those will be combined as text context. When multiple inputs include videos, the request can include up to 10 videos at once.
• --merge is ignored when any input is a video to ensure video references remain intact.