Raw footage in, rough cut out: the silence-cutting half of my pipeline

· 2026-09-21 · capcut, automation, editing

Everything I’ve written so far covers script-to-video automation: text in, finished short out. But the pipeline has a second mode for the opposite situation, when the video already exists and the editing is the problem.

What mode two does

Feed it raw filmed footage (a talking-head clip, a product demo, a voiceover take) and it:

  1. Detects silence and dead air and cuts it out.
  2. Detects stumbles and restarts in speech and drops them.
  3. Generates subtitles for what remains.
  4. Saves the result as a CapCut draft, so the fine-tuning happens on a rough cut instead of raw footage.

The philosophy is the same as the script mode: the machine does the mechanical first pass, a human makes the judgment calls.

The honest scope of this post

I’ll be straight about the receipts here: this mode is the newest part of my stack, and I have not yet run a formal before/after timing study on it the way I’ve measured the script pipeline. What I can state from my own history is the baseline it attacks: when I edited manually as a beginner, one video took me 40 to 60 minutes, and most of that time was exactly this, scrubbing for dead air and flubbed takes.

When I’ve put enough footage through it to publish honest numbers (minutes saved per raw minute of footage), that update will appear here. A receipt site doesn’t get to fake precision on its own tools.

Why silence-cutting is the right thing to automate

Cutting silence is the definition of judgment-free editing: nobody watches a pause and thinks “keep that.” It’s pure mechanical labor with an objective right answer, which is exactly the kind of work machines should take. What stays human is pacing, emphasis, and what to cut for story reasons. The rough cut just means you start from 80% instead of 0%.

What you can use today

You don’t need my pipeline for this concept:

  1. CapCut itself has a built-in “remove silence” style feature for simple cases; try it on one clip before building anything.
  2. If you record voiceovers, mark retakes out loud (“again”) while recording. Any silence-based cutter, mine included, works better when mistakes are followed by a pause.
  3. Measure your own baseline once: time one full manual edit. You can’t call any tool a time-saver without that number.

Here are the receipts.