This open-source repo does what OpusClip charges for, minus the watermark

Clipping long YouTube videos into vertical shorts is a solved problem commercially — OpusClip and Vidyo.ai both do it — but they cost money and add watermarks. This repo does the same job locally, for free, with no branding attached to the output.

The architecture is straightforward. faster-whisper runs locally to produce a timestamped transcript regardless of which LLM you choose. That transcript goes to an LLM — OpenAI, Gemini, or MuAPI — which first classifies the video by content type (podcast, interview, tutorial, vlog, and so on), then scores candidate moments against a virality framework: hook moments, emotional peaks, opinion bombs, revelations, conflict, quotables, story peaks, practical value. Each candidate gets a score from 0 to 100. Overlapping candidates are collapsed by score, and the top N are rendered as 9:16 crops.

The content-type classification step is the interesting design choice. Rather than applying a single generic highlight prompt to every video, the LLM tunes its criteria to the detected format and pacing. A tutorial and a podcast have different rhythms and different moments worth clipping; treating them identically would produce worse results. Whether the classification actually improves output quality in practice is something you'd have to test yourself — the README doesn't include benchmarks.

Gemini is the only provider with a free tier, though it carries a daily limit. OpenAI requires payment. MuAPI is pay-per-use without a subscription. Whisper runs locally across the board, so transcription costs nothing regardless of provider choice.

Beyond the CLI, there's a local web interface that lets you queue multiple videos and adjust flags visually. The .env file still handles credentials; the web layer only manages input and render options. That separation is clean.

The tool accepts local video files as well as YouTube URLs, which matters if you're working with footage you already have or content from platforms other than YouTube. Output goes to an output folder with no watermarks and no pre-clip credits — the README is explicit about this as a feature, positioning it directly against the SaaS alternatives.

For anyone building a content workflow that needs programmatic clip generation, the exposed API surface means this can be called from other projects rather than used only as a standalone script. That's a meaningful difference from a one-off utility.