tldr: Summarize Long-form Content

Summarize long-form content from YouTube (for now, more content support coming soon..). tldr, modified from TL;DR (Wikipedia), is a python library for summarizing. Originally tldr is used in context of texts, but this library can be used for summarizing any kind of data.

Installation

# Clone the repo
$ git clone https://github.com/nishantkumar1292/tldr.git
$ cd tldr

# Install dependencies (create a virtual environment and install dependencies in pyproject.toml file)
$ uv sync

# Install PyTorch with CUDA support (optional, for GPU acceleration)
$ uv pip install torch --index-url https://download.pytorch.org/whl/cu128

# Activate the virtual environment
$ source .venv/bin/activate

Usage

Python API (YouTube Example)

from tldr import YouTubeSummarizer

# Initialize the summarizer
summarizer = YouTubeSummarizer(
    model="o4-mini",  # OpenAI model to use - works better with reasoning models for long videos
    target_segments=7,    # Number of segments to create
    min_segment_minutes=3,  # Minimum segment duration
    max_segment_minutes=15  # Maximum segment duration
)

# Process a YouTube video
segments = summarizer.process("https://youtube.com/watch?v=YOUR_VIDEO_ID")

# Display results
for segment in segments:
    print(f"Title: {segment.title}")
    print(f"Summary: {segment.summary}")
    print(f"Duration: {segment.duration}")
    print(f"Time: {segment.start_time:.1f}s - {segment.end_time:.1f}s")
    print("-" * 30)

Note: You need to set the OPENAI_API_KEY environment variable for the summarization to work.

Segment Data Structure

Each segment returned by the API contains:

• start_time: Start time in seconds
• end_time: End time in seconds
• duration_minutes: Duration of the segment in minutes
• title: Descriptive title of the segment
• theme: Theme of the segment
• description: Detailed summary of the segment content
• text: Transcript of the segment

CLI

# Summarize a YouTube video
$ tldr youtube 'https://youtube.com/watch?v=YOUR_VIDEO_ID' --output-dir ./output

---

Content Types Supported

• Long youtube videos (LIVE) - Extract highlights from any video URL from YouTube, extracts the snippets, and also provides a summary of extracted snippets.
  • Customizable Params:
    • Number of segments to create
    • Minimum and Maximum duration of the segments
    • Model to use for the summarization (gpt-4.1, gpt-4-turbo, o4-mini etc.)
• Long Research Papers - Helps with extracting the key points from the paper. Creates a markdown file with screenshots of the relevant section, and creates a summary.

Extended scope

• Sports Highlights - This library can be used to create sports highlights from long sports match videos.

TODO

• Publish package to PyPI as omni-tldr
• Set up proper package structure with setup.py/pyproject.toml
• Create documentation and usage examples
• Implement research paper summarization functionality
• infer number of segments from the video length
• Build a web app to allow users to use this without setting up the environment
• Build an X handle to do this.
• cut video based on a prompt