What it is
A highlight cutter that keeps the reasons for every cut.
What it does: You give a video, its transcript and a target length, plus what you want to keep or drop. The tool builds candidate segments, shows exactly what text it will send, and after you allow the upload asks Jev to score each candidate. A planner then fits non-overlapping segments into the time budget, and FFmpeg renders separate clips and a merged reel with a review page.
Why it belongs in the showcase: Jev only scores text evidence; segment choice, timing and rendering stay in code, and the upload is opt-in.
Caveats: Rough cut only; no independent quality benchmark yet.
Evidence
Public MIT README (Chinese); reads SRT, VTT or Whisper-style JSON and keeps source timestamps; scores information value, opening hook, completeness, emotion and taste match; writes the full text request to `transfer-preview.json` first and uploads only with `--allow-text-upload`; outputs a local review page, separate MP4 cuts, the merged reel and a verification report; the core CLI uses only the Python standard library; tests use clearly labeled mock scores.
Caveats
- Outputs a rough cut only: no subtitles, dubbing, music or editor project export
- The author states there is no independent quality benchmark on long interviews, and the paid API was not re-run for the first open-source release
- You must hold the rights to the footage; transcript text goes to TypeSafe when scoring
Unofficial case page. Stars / list position observed for heat; do not treat heat as a quality SLA. Review the upstream README before you depend on it.