// the find
SamurAIGPT/AI-Youtube-Shorts-Generator
Open-source alternative to Opus Clip, Vidyo.ai, Klap & SubMagic. Turn long-form YouTube videos into viral 9:16 shorts using LLM highlight detection, Whisper transcription, and auto vertical cropping — free, no watermarks, no per-clip credits.
A CLI tool that clips long YouTube videos into vertical shorts by running Whisper transcription and an LLM virality-scoring pass on the transcript. The default mode routes everything through MuAPI, a third-party SaaS the same authors appear to run, so 'free and open source' comes with an asterisk. Local mode is genuinely offline except for the LLM highlight step.
The virality framework in highlights.py is concrete — hooks, opinion bombs, revelation moments, conflict — rather than generic 'interesting' scoring, which is a real improvement over naive approaches. Two-mode design (API vs local) is well thought out: same output shape, pluggable backends, and the local path actually works with faster-whisper + ffmpeg without phoning home. Long-video chunking with overlap is handled correctly so cross-boundary highlights aren't dropped. JSON output with full transcript and candidate metadata makes it composable with downstream pipelines.
The default mode is a thin wrapper around MuAPI, which is the repo author's own paid API — the README buries this and the 'free, no credits' framing is misleading since you're paying per-call to their service. OpenCV face tracking for vertical crop is brittle on anything without a clear face (screencasts, B-roll, talking-head-with-cutaways), and there's no fallback crop strategy. No evaluation harness: there's no way to measure whether the virality scores actually correlate with performance, so the whole ranking system is vibes-driven with no feedback loop. The batch processing story is just 'pipe URLs through xargs', which will serialize everything and has no error recovery if one video fails mid-batch.