Jev 101
Start Guides
Search
中文 EN
Media outlineExperimental — Jev scores transcript text only; rendering runs locally with FFmpeg

Transcript highlight select

Who: Editors cutting long interviews, podcasts or streams. Steps: Split the transcript into timestamped candidate segments; ask Jev Score and Noul questions for information value, opening hook, completeness and fit with a written taste; pick non-overlapping segments within the target length in code, have a person review, then render locally. Expected effect: a rough cut with a traceable reason for every segment.

Source: https://github.com/gnipbao/jev-highlight-cutter ↗ Docs / source ↗
1

Treat this as an outline — adapt state and questions to your data.

2

Implement in code — call System One / Jev; compose answers yourself.

3

Gate on confidence — act, confirm, or escalate before side effects.

Outline sketch

Prompt 
# Transcript highlight select
State: one candidate transcript segment with its timestamps, plus the editor's written taste.
Ask: Score for information value, opening hook and completeness; Noul for whether it matches the taste and whether it is an ad or a repeat.
Then: rank in code, pick non-overlapping segments within the time budget, and let a person review before rendering.

Needs access to: ffmpegwhisper

Who it's for

Editors cutting long interviews, podcasts or streams

Steps / how it's set up

Steps (experimental)

  1. Build candidate segments from the transcript and keep the original timestamps.
  2. Ask Jev a few Score / Noul questions per segment, using a written description of what to keep and what to drop.
  3. Pick segments in code within the time budget, review them, then render locally.

Sources

This pattern is experimental. No independent quality benchmark yet — review every cut yourself.

Expected effect

Taste-aware highlight picks with a reason for every cut

Unofficial outline for learning. Paraphrased from public docs and tutorials — not a production recipe. Review sources before you automate anything.