A single one-hour podcast episode contains, on average, between fifteen and twenty-five moments that work as standalone short-form clips. The bottleneck was never the raw material — it was the hours of manual editing needed to turn each moment into something worth posting. Style cloning removes that bottleneck by separating two things editors normally do at once: choosing what to show, and deciding how it should look.
What a reference video actually encodes
When you watch a creator whose edits you admire, you are responding to a set of measurable, repeatable decisions rather than to taste in the abstract. Average shot length. How long a speaker stays on screen before the frame pushes in. Whether captions sit at the vertical centre or ride two-thirds down the frame. How aggressively the audio is compressed. Whether cuts land on the beat or slightly ahead of it.
Each of those decisions leaves a signature in the file. Cut frequency is visible in the frame-difference histogram. Zoom behaviour shows up as a scale ramp over a known number of frames. Caption styling is a font, a weight, a stroke width, and a vertical offset. None of it is mysterious once you stop treating "editing style" as a single unanalysable quality.
Extraction, not imitation
The distinction matters. Imitation would mean copying a reference video's content. Extraction means measuring its parameters and applying those numbers to entirely different footage. What transfers is the rule set — cut every 1.8 seconds on average, punch in 12% on emphasis, captions in a heavy sans at 68% frame height — not a single frame of the original.
This is why the output does not look like a copy of the reference. It looks like your footage, edited by someone who works the way the reference editor works. The subject, the speaker, the words and the B-roll are all yours.
Where the twenty clips come from
Transcript analysis identifies self-contained segments: a question and its answer, a story with a beginning and an end, a claim followed by its justification. Segments that depend on earlier context get dropped, because a clip that assumes knowledge the scroller does not have will lose them in the first two seconds.
From a sixty-minute conversation that typically leaves eighteen to twenty-two viable segments. Each is then cut according to the extracted style profile, reframed to 9:16, captioned, and exported.
- Segments run 20–60 seconds, the window where retention is measurable on all three major platforms
- Each clip opens on the strongest sentence in the segment, not on the chronological beginning
- Speaker changes drive reframing so a two-person conversation stays legible vertically
What still needs a human
Style cloning is good at consistency and bad at judgment. It cannot tell you which of the twenty clips is the one worth putting money behind, and it will occasionally select a segment that reads badly out of context. Review before publishing is not a formality — it is where the value of the remaining human step sits.
Try it on your own footage
Thirty processing minutes a month on the free plan. No credit card required.
Get early access