Video Summarizer using Local LLM (facebook/bart-large-cnn)

Submit a YouTube URL and get an AI-generated summary of the video.
Explore the code »

View Demo · Report Bug · Request Feature

This project allows you to generate AI-powered summaries of YouTube videos using the facebook/bart-large-cnn local language model. You just send a video URL to the FastAPI endpoint, and within ~8 seconds (depending on your GPU), you get a clean summary of the content.

💡 It works even for long transcripts or complex documents.

📃 Added Gradio Frontend ⚙️ You can now select summary length:

• very short
• short
• normal
• long
• very long

1. Transcript download using get_transcript.
2. Chunking of the text for better LLM input.
3. Chunk-wise summarization with facebook/bart-large-cnn.
4. Merging summaries into a final summary.
5. Exposed via API endpoint:
   GET /summarize-video?url=YOUR_YOUTUBE_URL

• ai_core.py: Handles tokenization and LLM model logic
• ai_algorithm.py: Chunking and hierarchical summarization logic
• get_transcript.py: Retrieves transcript from YouTube
• api.py: FastAPI web server exposing the summarization endpoint
• tests/: Unit tests for core components

• 
• FastAPI
• Hugging Face Transformers

This repo contains a full suite of unit tests for the AI pipeline logic to ensure consistency and reliability.

Run them using:

python -m Backend.tests.unit_test

This project is licensed under CC BY-NC 4.0. That means:

• ✅ Free for non-commercial use
• ✅ Attribution required (mention the author)
• ❌ Commercial use is forbidden

Author: Mikołaj Jaros. (aka MixDevv)

• Hugging Face
• FastAPI
• Python Software Foundation