Skip to main content
splice edits video by editing a JSON document. You make typed changes — trim, retime, caption, composite, grade — and splice compiles the result into ffmpeg argv, runs it, and caches each step. Every command answers with exactly one JSON object, so a program (or a person reading a transcript) can tell what happened without scraping logs.

​
Why it exists

Raw ffmpeg is a filter-graph language: correct, but hard to review, hard to version, and easy to get subtly wrong. splice keeps the edit in splice.json, which you can diff, hand to a reviewer, or regenerate from a script.

Quickstart

From nothing to a rendered file in five commands.

Install

npm, the ffmpeg requirement, and what to do when a system build is broken.

Command reference

Every verb, with the flags that matter.

For agents

The loop, the contract, and the anti-patterns.

​
What it is good at

  • Cutting and retiming a recording, including removing fillers and long silences.
  • Captions: transcribe, review the SRT, burn it in as two separate decisions.
  • Compositing: layers, a moving camera, sliding placement, masks, one grade.
  • Iterating cheaply: a preview profile, a per-step cache, and plan to see the argv before spending time.
  • Being driven by an agent: one JSON object per command, warnings that say what will not do what you asked, and errors that name the command that clears them.

​
What it is not

  • A transcoder. If the job is “change container” or “re-encode at another bitrate”, use ffmpeg directly.
  • A timeline GUI. The document is the interface.
  • A streaming or broadcast tool. It renders files.

​
Requirements

node 18+ and ffmpeg with libx264 and libass. Linux builds are static, so Alpine works too. splice doctor is the first command to run anywhere new: it reports which ffmpeg won, which encoders and filters exist, and whether a font was found.