Focus, create, grow: editing talking-head video for vertical without losing the plot

Talking-head video is the easiest format to record and the easiest to ruin in the edit. The camera doesn't move, the framing doesn't change — so every second of dead air is felt. Here's the lean checklist we see working across thousands of clips.

Cut the first answer, keep the second

People warm up as they speak. The first phrasing of an idea is usually hedged; the restatement thirty seconds later is the quotable one. When clipping, search for the sentence that would work as a screenshot — start there, not at the start of the topic.

The 10-second rule

Something must change every ~10 seconds: a caption emphasis, a zoom punch, a b-roll flash, a beat of silence removed. None of these require reshooting — they're all edit-layer moves. Viewers don't consciously notice them; they notice their absence when their thumb starts moving.

Reframe for faces, not for frames

Cropping 16:9 to 9:16 with a static center crop beheads speakers the moment they lean. Face-tracked reframing follows the active speaker and keeps eyelines in the upper third — which is where vertical viewers look first. If there are two speakers, cut between them instead of squeezing both into frame.

Silence is an edit, not a flaw

• Strip filler words and long pauses — a 58-second raw moment usually plays better at 41 seconds.
• But keep one deliberate pause before the punchline; contrast sells it.
• Snap cut points to word boundaries so the audio never clips mid-syllable.

Focus the moment, create the rhythm, grow from the data. Then go record the next one — the edit is no longer the bottleneck.

The anatomy of a talking-head clip that holds

Break down any high-retention talking-head Short and you'll find the same skeleton: a hook delivered in the first line of on-screen text (not just speech), a single idea developed without detours, visible rhythm changes on a roughly 10-second cadence, and an ending that lands within two seconds of the last useful word. Every editing decision below serves one of those four beats — if an edit doesn't, cut the edit.

Zoom punches: the cheapest rhythm tool

A zoom punch — cutting from a full frame to roughly 110–120% scale on the same shot — costs nothing and reads as an edit event. Place them on emphasis words, not on a timer: the punch should agree with the speaker's own stress. Two rules keep them classy: never punch twice in a row in the same direction, and never punch during a caption emphasis — pick one attention cue per moment.

B-roll without a b-roll budget

• Screenshots beat stock. If the speaker mentions a tweet, a chart, or an app — show the actual thing for 1–2 seconds.
• Keyword-matched stock works for concepts: 3–5 seconds max, and always under the speaker's continuing audio, never replacing it.
• The speaker's own hands, notes, or screen make the most authentic cutaways of all — shoot 30 seconds of them after every recording.

Audio is half the retention

Viewers forgive soft visuals; they do not forgive muddy audio. Before any visual polish: remove background music bleed and room noise, then restore voice clarity. AI enhancement chains (music separation followed by speech restoration) rescue recordings that would otherwise be unusable — a phone recording in a café can come out sounding like a booth. If a clip's audio can't be rescued, kill the clip; no caption style saves unintelligible speech.

The last step is the loudness check: normalize so your voice sits at a consistent level across every clip you publish. Feeds are consumed in sequence — if your clip is quieter than the previous video, the thumb moves before the brain even engages.