AI-Youtube-Shorts-Generator Hits 4,200 Stars Replacing $300 Clipping Tools
An open-source clipping pipeline uses LLM highlight detection, Whisper, and auto vertical cropping to turn long videos into ranked 9:16 shorts.
Takeaways
- Open-source Python tool turns long YouTube videos into ranked 9:16 shorts via LLM highlight detection.
- Two modes: API-based using MuAPI, or fully local with yt-dlp, faster-whisper, and OpenAI or Gemini.
- Ranks clips on hooks, emotional peaks, opinion bombs, revelations, conflict, quotables, and practical value.
- Auto-chunks videos over 30 minutes into 20-minute overlapping segments so cross-boundary highlights are not missed.
- Ships CLI, Python library, JSON output, batch processing, and configurable virality framework in
highlights.py. - MIT licensed with 4.2k stars, positioned as free alternative to Opus Clip, Vidyo.ai, Klap, and SubMagic.
A Python repo called AI-Youtube-Shorts-Generator has crossed 4,200 stars by positioning itself as a self-hostable stand-in for paid clipping tools like Opus Clip, Vidyo.ai, Klap, and SubMagic. Point it at a YouTube URL and it returns a ranked set of vertical shorts, each with a viral score, an opening hook, and a one-line explanation of why the clip should perform.
The pitch targets people who do not want to pay $20 to $300 per month or hit caps on minutes processed: no per-clip credits, no watermarks, and full control over the highlight algorithm. Under the hood it stitches together a fairly conventional stack of yt-dlp for downloads, Whisper for transcription, an LLM for highlight ranking, and FFmpeg plus OpenCV for the vertical crop.
Two paths through the pipeline
The project ships with two execution modes that share the same output shape. The default API mode uses MuAPI for download, transcription, and cropping. Local mode runs entirely on your machine with yt-dlp, faster-whisper, and ffmpeg/opencv, and lets you pick OpenAI or Gemini for highlight ranking.
Local mode is the interesting one if you already have GPUs sitting idle. Whisper transcription can run in the cloud through MuAPI or locally via faster-whisper on CPU or CUDA, so only the LLM call has to leave your machine. The tradeoff is setup: you need