Does Reelsaw handle video podcasts with two cameras?

Yes. With two people in a wide shot it follows whoever is talking; when detection is not confident it stacks both speakers in a split layout. Each clip is split into sections at scene cuts, and you can force any section to face, split, stack, fit or centre.

Does it work on audio-only podcasts?

It needs a video input. For audio-only shows, render a static or waveform video first; the fit layout places it over a blurred background with captions and a hook.

Can my producer and I both use the same install?

Yes. The first account is the admin, sign-up can be open, approval-based or closed, and each person gets their own project defaults, vocabulary, API keys and usage. Admins can set per-user quotas.

Guide

Podcast to shorts without paying by the minute

Podcasts are the worst case for per-minute pricing: long episodes, two or three faces, and a lot of 'um' between the good parts. Reelsaw was built around that shape: it chunks long transcripts without losing the middle, stacks speakers when it cannot pick one, and cuts fillers with cuts it has checked.

Long episodes without losing the middle

Language models rank what they see at the start and end of a long prompt better than what is in the middle. Reelsaw chunks the transcript at long pauses with 20% overlap, collects every standalone moment per chunk in parallel, then pools and ranks them. Weak picks are dropped relative to the run, not by a fixed count, so a two-hour episode yields as many clips as it earns.

Two people in frame

Split: two speakers stacked top and bottom, when the shot has two faces and detection is confident.
Face: the crop follows whoever is talking, chosen by mouth motion correlated with the audio envelope, with a minimum of 2.5 seconds per shot so it never flickers.
Stack: slides or a screen on top, a zoomed webcam below, for video podcasts that share a screen.
Fit: the whole frame over a blurred background when there is no face, such as a title card or b-roll.
Clips are split at scene cuts and face-presence changes, so one clip can open on a wide shot and switch to a talking head. Override any section in the editor.

Fillers, repetitions and false starts

Disfluencies are cut with jump cuts. Each cut is located by forced alignment and checked by listening to the seam, so karaoke highlights still land on the syllable. Every fix shows inline in the transcript with accept and reject, long pauses can be shortened, and your decisions survive re-runs. A wrong automatic edit is worse than a missed one, so anything uncertain goes to review instead of being applied.

Names and jargon

Guest names, product names and recurring terms go into a vocabulary. Transcription is biased toward them and a constrained correction pass fixes the rest, so captions do not misspell the person you just introduced.

What a two-hour episode costs

What a two-hour episode costs
	Reelsaw	Hosted, metered tools
Software	$0	Plan with a monthly minute or credit budget
120 minutes of source	Local: electricity. Cloud transcription: that vendor's per-minute price, no markup	120 credits or minutes from the budget; several episodes a month usually means the next tier
Where the episode goes	Your machine	The vendor's servers
Back catalogue of 100 episodes	Queue them; pg-boss runs them through and resumes after restarts	Buy credits or spread it over months

Common questions

Does Reelsaw handle video podcasts with two cameras?: Yes. With two people in a wide shot it follows whoever is talking; when detection is not confident it stacks both speakers in a split layout. Each clip is split into sections at scene cuts, and you can force any section to face, split, stack, fit or centre.
Does it work on audio-only podcasts?: It needs a video input. For audio-only shows, render a static or waveform video first; the fit layout places it over a blurred background with captions and a hook.
Can my producer and I both use the same install?: Yes. The first account is the admin, sign-up can be open, approval-based or closed, and each person gets their own project defaults, vocabulary, API keys and usage. Admins can set per-user quotas.

Sources

Last updated 2026-10-18.