Skip to content

jev-shorts-editor

Turns a talking-head clip into a finished vertical short, with every edit decision made by Jev through a LiteLLM gateway.

Projectprototype

Overview

jev-shorts-editor turns a raw talking-head video, retakes and dead air included, into a finished 1080x1920 vertical short with one command. It cuts the clip, reframes the speaker, and adds animated captions, graphic cards, zooms, transitions and sound effects. Every edit decision is a Jev call sent through the /v1/decisions route of a LiteLLM gateway, which holds the TypeSafe key and adds virtual keys, budgets, spend tracking and request logs.

Jev decides which repeated lines are real retakes, a per-video style (style family, accent color, caption style, opening title, energy, progress bar), caption style roughly every 8 seconds, and for each 2 to 3 second shot the camera move, overlay, card template, text effect, transition, sound effect and word to emphasize. Code turns answers into a plan, with rhythm rules that keep something changing at least every 3 seconds and log overrides. The only generated text is card copy from a small LLM on Groq. On one 4.5 minute take cut to 42 seconds, 22 parallel requests finished in 0.51 s for $0.0023. It runs on macOS 15+ on Apple Silicon and installs as an agent skill.

What the decision model decides

TypeQuestionAnswers
Yes / noIs this repeated line a real retake to cut?(inferred)yes · no
ChoiceHow should the camera treat the speaker during the current shot?base · punch · push_in · face_closeup · drift
ChoiceWhich transition fits this cut?(inferred)—

How it works

  1. Install the skill and give your agent a raw clip
  2. Transcribe and clean retakes and dead air
  3. Jev picks per-video style and per-shot edits in parallel
  4. Code builds a plan with rhythm rules
  5. Render the 1080x1920 MP4 and check it

Strengths

  • Every edit is a bounded choice, nothing to parse
  • Spend and logs tracked through LiteLLM
  • Very low decision cost per video

Caveats

  • macOS 15+ on Apple Silicon only
  • LiteLLM decisions route needs a source install for now
  • Needs TypeSafe and Groq keys plus ffmpeg

Getting started

npx skills add misbahsy/jev-shorts-editor, then ask your agent to make a short from a video file

At a glance

Type
Project
Status
prototype
Language
TypeScript
Runtime
Node CLI and agent skill
Has code
yes

Use cases and integrations

How curated lists describe it

Frequently asked questions

What does Jev decide?

Retake cuts, the overall style, and per-shot camera, overlay, card, text effect, transition, sound and emphasis choices.

Does Jev write any text?

No. Only card copy is generated, by a small LLM on Groq.

What did a real run cost?

22 edit requests for a 42 second short cost $0.0023 and finished in 0.51 s.

Related projects

  • Video summarizer that ties every line to its second; Jev checks each summary sentence against what was said, with numbers checked in code.

    ProjectbetaPython
  • Claude 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
  • Codex skill and Python CLI for local video remixes with shot and beat detection, Jev clip judgments and an originality report.

    Agent skillprototypePython
  • HeyGen post on a lead pipeline where Jev decides in milliseconds which leads deserve a personalised video before rendering.

    Discussiondiscussion
  • IDE and browser extensions to find and install agent Skills and Subagents, with an opt-in Jev check when local detection can't decide.

    Browser extensionproductionTypeScript
  • Reusable project context and specialist workflows for existing AI tools, with an addon that offloads bounded decisions to Jev.

    ProjectproductionTypeScript