Jump cut

A jump cut removes time from inside one continuous shot, so the same subject in the same framing jumps forward — the talking-head internet's default cut.

What actually changes across the cut

There are two kinds, and they are usually discussed as one. A temporal jump cut removes a middle section from a single continuously filmed shot. A spatial jump cut joins two shots that were never continuous but happen to share framing. Almost every jump cut in online video is the first kind, and it is the one silence removal manufactures.

The reason a temporal jump cut is visible has nothing to do with the cut itself and everything to do with what the frames either side of it are carrying. At 30 fps, the last frame before the cut and the first frame after are 1/30th of a second apart on screen and one to three seconds apart in real time. In that gap the speaker breathed, shifted weight, moved their hands out of frame, blinked, and rotated their head. Research on smoothing these cuts measures exactly that: how far the head and body pose travels across the junction, because the size of the pose change is what decides how hard the cut hits. The picture is continuous in every dimension the eye uses to establish that this is one shot (background, lighting, lens, colour) and discontinuous in the one dimension that is moving. That mismatch is the jolt.

Classical continuity editing solves this by refusing the setup. The 30-degree rule says that two successive shots of the same subject must be at least 30 degrees apart around it; a variant, sometimes called the 20mm/30-degree rule, adds that a focal-length change under about 20mm at the same angle fails for the same reason. Move far enough and the audience reads the cut as a new view of the same scene rather than a skip in time. A significant change in shot size — medium to close-up — is the accepted substitute. Every one of those remedies requires a second camera position, which a single locked-off webcam does not have.

Why silence removal produces jump cuts by definition

A silence remover is a jump-cut generator, and this is not a flaw in any particular implementation. The operation is: find spans in a continuous take where nobody is talking, and delete them. Deleting time from inside a continuous take is the definition of a temporal jump cut. The camera cannot have moved during the gap, because there was no gap in the recording — only in the speech.

The parameters that control the operation control how many you get. Every silence detector has three:

• A threshold — how quiet counts as quiet. Waveform-based detectors sit somewhere around −35 to −40 dBFS. Set it too high and the tool eats word onsets and room tone; too low and a noisy room never registers a pause at all, which is why speech-gap detection (using the transcript rather than the waveform) is more reliable on footage with a music bed or background noise.
• A minimum gap length — typically 0.5 to 0.75 seconds. Below that you are cutting the natural rhythm of a sentence, not dead air.
• Padding — how much silence to leave on each side, commonly 0.1 to 0.25 seconds. This is the single setting most responsible for whether the result sounds edited. Zero padding clips the breath before a word and the decay after it; the picture jolt then arrives at the same instant as an audio artefact, and viewers who could not name either will say the video "feels cheap".

Run those numbers over a real recording and the count is large. A 3.9-minute talking-head video edited in Valmera produced 45 cut junctions — about one every five seconds — and essentially all of them sat inside a single unbroken shot. That number is the whole problem in one figure: any per-cut treatment you might reach for now has to survive being applied 45 times.

The four responses to the jolt

A scenario you will actually hit

You record a 22-minute unscripted explainer on one camera, no teleprompter. Roughly a third of the runtime is thinking pauses, restarts and "um". You strip the dead air and the fillers and land at 14 minutes, which is a genuinely better video — and about 160 jump cuts.

The mistake at this point is to treat all 160 as one problem with one answer. They are not equal. Most are half-second pauses where you barely moved, and they are already invisible. The ones that read badly are the handful where you removed a long pause and your hands, head or posture teleport — the biggest pose changes. So the work is: watch it once, mark the four or five that snag, punch in or drop coverage on those, and leave the rest alone. Finding them takes less time than applying anything globally, and applying anything globally is what produces the whip-pan-every-two-seconds edit.

The second mistake is fixing the picture and ignoring the audio. If the cuts were made with no padding, the video still feels wrong after every visual fix, because the artefact is a clipped breath rather than a skipped frame. Re-cut with padding before you reach for punch-ins.

Common mistakes and misconceptions

How this works in Valmera

Valmera cuts silence in one request, defaulting to a 0.5-second minimum gap and 0.12 seconds of padding around speech, with every cut snapped to a word boundary from the word-level transcript rather than to a waveform crossing. Gaps are found in the speech, not the waveform, so a recording over a music bed still gets cut correctly. Cuts are entries in an edit decision list, so the original file is untouched and anything cut can be restored by asking.

The part specific to jump cuts is transitions. Valmera separates a scene change from a jump cut inside one shot using shot detection, and places transitions only at real scene changes by default. If every junction is a jump cut — the normal state of a silence-cut talking head — it does not apply the transition and says why. One on every cut is available, but has to be asked for. That behaviour exists because a real user received a preview with 45 whip pans through a single continuous shot and correctly called it broken.

For the other fixes: zooms can be placed over any range of the finished program and aimed anywhere in the frame, and an automatic punch-in pass targets the most vocally stressed words — three zooms at 14 percent by default. The honest limitation is that this pass aims at emphasis, not at cut junctions; for a punch-in on a specific jump cut, describe that moment. B-roll is an insert placed at a timestamp, from your own upload, a pasted link, stock, or a generated still or clip.

What Valmera does not do here. No true crossfade or dissolve — the seven transition styles are all duration-preserving junction effects, and one style applies to the whole video rather than being chosen per cut. No AI frame synthesis: Valmera never fabricates frames between two real ones. No motion tracking, so coverage cannot follow a moving subject. And no way to remove silence without producing jump cuts, here or anywhere else, because that is what the operation is.

Related terms

• Silence detection — The mechanism that generates jump cuts — thresholds, minimum gap length and padding.
• B-roll — The only fix that removes the discontinuity rather than masking it.
• J-cut and L-cut — Split audio/picture edits. Routinely confused with jump cuts; unrelated.
• Word-level timestamps — Why a machine cut can land on a word boundary instead of mid-syllable.
• Shot detection — How a tool tells a real scene change apart from a jump cut inside one shot.
• Edit decision list — The list a cut is recorded in, which is why a cut can be undone.
• Text-based video editing — Deleting words from a transcript — a jump-cut generator by construction.

See the full video editing glossary, or the tool pages for removing silence, removing filler words and transitions.

Frequently Asked Questions

Edit a talking head without touching a timeline

Upload, say "cut the dead air and the ums, punch in on the emphasis", watch the preview. 50 free credits, no card.