Universal Video Summarizer

Universal Video Summarizer is a python based QT5 desktop application that automates summarizing videos. The application utilizes yt-dlp to download videos, then ASR to transcribe it and LLMs to summarize/process the transcript according to user directives.

Installation

Prerequisites

• Python 3.8+
• ffmpeg installed on your system.

Install Dependencies

pip install -r requirements.txt

Usage

Running the Application

python main.py

Using the Application

1. Enter the URL or local path of the video you want to summarize.
2. Select the desired summarization method.
3. Click the Summarize button to start the summarization process.
4. The application will display the summarized text in the text box.

Selecting Which Model to Use

1. Click the Settings button.
2. Select one of many LLM models available on Hugging Face.
3. Click the Save button to save the settings.

Recommended models:

ASR Models

Model Name	Notes
openai/whisper-large-v3	Best performance.
distil-whisper/distil-large-v3	Reduced VRAM usage, comparable performance.

LLM Models

Model Name	Notes
meta-llama/Llama-3.2-1B-Instruct	Requires signing up and requesting access.
meta-llama/Llama-3.2-3B-Instruct	Larger version of the 1B, requires more VRAM to run fast. Requires signing up and requesting access.
Qwen/Qwen2-1.5B-Instruct	Has issues with hallucinations.

Features

• Summarize videos from YouTube, Twitch or any other online platform yt-dlp supports and local files.
• Switch between different models for transcription and summarization.
• Create custom prompts for the summarization process.

Contributions

Contributions are welcome! Feel free to fork this repository, submit issues, or create pull requests.