Podcast clips: from a long episode to vertical shorts
This workflow takes one long podcast episode and turns it into vertical shorts: choose highlights from the transcript, reframe to vertical, add captions and export. Unlike tool reviews, it counts the human finishing time, so you see what a clip really costs to produce. Prompts and files are shared.
This page is built but kept out of search until it carries the evidence it needs: Needs a recorded podcast we may use; in testing.
The result
We ran this job on three 60-second film clips with free tools. Each run took about 22 to 28 seconds on a CPU-only laptop and produced a 1080x1920 file with three-word captions. Full numbers, the script and what failed are on We cut a vertical short out of three 60-second clips and measured the result.
What we did not test: a real podcast with several speakers, a long recording, a face-tracking crop, or any paid clipping service. Our clips are film excerpts.
What failed or surprised us
- Picking the stretch by speech density found dense talk, not good moments, and one pick started mid-sentence.
- Single-pass loudness missed its -14 LUFS target on all three files. Two passes got two of three within 0.1; the third was 0.4 over.
- A blurred-background layout leaves a thin strip of picture in the middle of a vertical frame.
The steps
- Transcribe the whole video with word times (faster-whisper small.en).
- Pick a 15 to 25 second stretch of whole transcript segments with the most words per second, and check it by watching it.
- Cut it, build the vertical layout, burn in captions of at most three words, level the loudness: one FFmpeg command.
- Check the captions against the safe zones with Shorts, Reels and TikTok safe zone overlay before you post.
What it took on this laptop
Median runs of 21.9, 28.4 and 24.6 seconds for a 60-second clip, of which transcription was about 12, 14 and 13 seconds. An AMD Ryzen 7 7735U with no GPU in use. Transcription time grows with the length of the recording.
Method and limits
- Tested on film excerpts, not on a podcast.
- A person should watch every short before it goes out.
- Our method is on How we curate and test AI video workflows.
The workflow is run on one long recording with tool versions stated. A clip counts as accepted only when a person finds no cut-off words, no caption errors and no faces outside the frame. The time a person spends reaching that point is recorded.
- Highlight quality depends on the episode, so results from one show do not transfer to every podcast.
- Human finishing time varies with your standards and how fast you work.
- Check platform specs on the safe zone checker, because platforms change them.
Page checked on Nov 4, 2026.