jevclip
CLI that has Jev judge a video's subtitles segment by segment, then cuts a highlight reel and a filler-free full version.
Overview
jevclip keeps the valuable parts of a video. Jev judges subtitles or a transcript segment by segment, and code uses ffmpeg to cut two videos: a roughly three-minute highlight reel and a de-watered full version that removes only segments that are clearly useless. Greetings, ads, empty talk, motivational filler and exaggerated claims are filtered with a written reason for each drop; kept segments are ranked by value and joined in original order for the highlight. The full version errs on the side of keeping information. An optional focus argument keeps only segments relevant to a topic, and transcripts alone produce selections and summaries without video.
Summaries, written by an optional OpenAI-compatible model (or falling back to excerpts), link each sentence back to a timestamp. The README stresses that default thresholds are placeholders and provides export and eval commands to label about 50 segments and pick the smallest threshold meeting a maximum error. On a synthetic 9-minute demo with 12 pre-labelled blocks, it reports every block judged correctly and 12 Jev requests costing about $0.0007. It uses only the Python standard library plus ffmpeg, caches judgments locally, and the README is in Chinese.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this segment filler (greeting, ad, empty talk, exaggeration)?(inferred) | — |
| Score | How valuable is this segment?(inferred) | — |
| Yes / no | Is this segment relevant to the requested focus?(inferred) | — |
How it works
- Split subtitles or transcript into segments
- Jev judges each segment's value and filler status
- Code cuts a highlight reel and a filler-free full version with ffmpeg
- Write a summary linked to timestamps; record reasons for drops
Strengths
- Every drop has a stated reason
- Tools to calibrate thresholds on your own labels
- Very low cost per video per the README's demo
Caveats
- Default thresholds are placeholders
- Demo evaluation is synthetic
- Chinese documentation
Getting started
export TYPESAFE_API_KEY=..., then `jevclip run video_dir/` (install with `uv tool install .`).At a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- CLI + ffmpeg
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Jev-powered video highlights and cited summaries from subtitles and scripts.
hellogumbo/awesome-jevJudges a video's transcript segment by segment and cuts two versions, a short highlight reel and a full one with the filler removed, writing down why each cut was dropped. Chinese.
MrJev/awesome-jev
Frequently asked questions
What two videos does it produce?
A short highlight reel and a full version with only clear filler removed.
How should I set thresholds?
Label about 50 segments and use `jevclip eval` to pick the threshold meeting your error target.
Can I run it on a transcript only?
Yes, it then produces selections and a summary without cutting video.
Related projects
Record linkage for researchers: write the match rule in plain English and Jev judges each candidate pair with a probability.
ProjectbetaPythonComfyUI custom nodes that use Jev to select candidates, judge conditions, score text, extract numbers and choose local Skills.
ServiceprototypePythonJSON-in/JSON-out CLI that reranks, filters or compresses search results with independent Jev Noul checks, keeping metadata intact.
CLIbetaRustScripts where Jev selects CVSS metric values from a vulnerability description and code computes v3.0, v3.1 or v4.0 scores.
ProjectbetaPythonSite audit for SEO and AI-search visibility: a deterministic crawl, narrow Jev Choice, Score and Noul questions, and a live dashboard.
AppbetaTypeScriptPython library that replaces hard-coded RAG thresholds with calibrated decisions for stopping, chunking, selection, abstention and caching.
ServicebetaPython