jevcut
Auto-clipper for talky videos: code enumerates every possible cut point and Jev judges where each standalone clip starts and ends.
Overview
jevcut turns a long video such as a podcast, talk, interview or comedy set into short clips that stand on their own for Shorts, Reels or TikTok. One command transcribes the video, finds moments worth clipping, decides exactly where each starts and stops, drops anything that does not stand alone or turns out to be an ad, and writes ranked mp4s plus a page to watch them. The design bet is that auto-clippers usually fail at the edges, starting mid-thought or before the punchline lands. So code lists every place a clip could start or end (sentence ends, pauses, speaker changes) and never lets a model name a timestamp; Jev only judges the clip a candidate edge would produce or picks among enumerated candidates.
On 38 hand-labelled videos across 13 genres, the README reports jevcut beating five simpler approaches on recall, on how often both edges match a human cut, and on precision, with full numbers and reasoning in separate documents. On a pilot set it used about 520 Jev requests per hour of video, roughly $0.03, with transcription costing about five times more. Version 1 handles only verbal content, since moments without words are invisible to a transcript-based judge. MIT licensed.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Does the clip this candidate edge would produce stand alone?(inferred) | — |
| Choice | Which enumerated start or end point makes the best clip edge? | — |
| Yes / no | Is this moment an ad? | — |
How it works
- Transcribe the video
- Scan for moments worth clipping
- Code enumerates candidate edges; Jev judges or picks among them
- Drop non-standalone or ad clips; render ranked mp4s
Strengths
- Benchmarked on 38 hand-labelled videos
- About $0.03 of Jev per hour of video per the README
- Model never invents timestamps
Caveats
- Verbal content only in v1
- Transcription is an extra cost
Getting started
uv run jevcut run talk.mp4 --model lemonfoxAt a glance
- Type
- Benchmark
- Status
- beta
- Language
- Python
- Runtime
- CLI
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Auto-clipper that turns long videos (podcasts, talks, essays, comedy) into short standalone clips for Shorts, Reels and TikTok. Code lists every possible cut; an AI judge picks where each clip starts and ends. Benchmarked on 38 hand-labelled videos.
logicrw/awesome-jev-projects
Frequently asked questions
Can the model choose any timestamp?
No, code enumerates candidate edges and the model only judges or picks among them.
How much does it cost?
About $0.03 of Jev per hour of video, with transcription costing more.
What content does it handle?
Talk-heavy video; moments without words are out of scope in v1.
Related projects
Experiment testing whether Jev knows music theory via SATB part-writing and quizzes vs Claude, plus a Jev-driven Rhodes piano demo.
AppresearchJavaScriptVideo demo classifying 50 mock YouTube comments with Jev (intent, value, video idea) and comparing speed and cost with GPT-4o-mini.
ModelprototypePython- jev-rerank★ 17
Jev as a calibrated reranker: query plus up to 30 documents in one call, one relevance Noul per document, benchmarked on BEIR.
ProjectbetaPython CLI that has Jev judge a video's subtitles segment by segment, then cuts a highlight reel and a filler-free full version.
ProjectbetaPythonAgentic LEGO builder that turns an idea into an LDraw model, using Jev reranking to find suitable parts and example models.
ProjectbetaPythonClaude skill and API that turns a video link into timestamped JSON, then asks Jev typed questions to rank or filter videos against a brief.
Agent skillbetaPython