Long-Form Video Clip Pipeline

To automate repurposing long YouTube videos into short clips, you can run a pipeline that downloads the video, transcribes it locally with Whisper, and uses Claude to select viral-worthy segments. This end-to-end process automates download, transcription, AI segmentation, and FFmpeg cutting.

AI segmentation evaluates hook strength and narrative completeness to select the best sections for content repurposing. It uses Claude to choose 3–5 standalone segments, verifies cut endpoints with an LLM check, and optionally scores candidates with a 10-expert rubric to output only high-quality clips.

Yes, you need FFmpeg and yt-dlp installed locally to cut video clips from YouTube. The pipeline requires yt-dlp for downloading source episodes and FFmpeg for accurately cutting the selected highlight clips, along with local Whisper for transcription.

The best way to scale back-catalog video clipping for a podcast is using an automated pipeline with a scored winners-only mode. This approach processes many episodes into shareable assets, cutting only segments above a minimum quality threshold to maintain consistent clip selection over time.

Yes, you can use OpenAI Whisper for local transcription of YouTube videos within this pipeline. Whisper handles the local transcription of downloaded audio, which is then passed to the Claude API for AI-based segment selection and cutting verification.