Two-second lines on fast cuts

Jump cuts kill a caption hold. A two-second line cannot sit on a half-second shot. The words stay while the picture has already changed, or the next cue arrives on a face that is not speaking. Re-time after the edit. Do not caption the long take and then cut it to pieces.

The hold is a duration, not a vibe

Broadcast practice parked cues for about one to six seconds: long enough to read, short enough to yield. Social inherited the two-line block and then invented jump-cut talking-heads. The mouth pops forward in time. Dead air is gone. Energy is up. The caption clock is still the old waveform.

Caption timing is when the block appears and when it leaves. Reading speed is whether a human can finish it while it is there. Fast cuts are a third clock: how long the shot exists. If the shot is 12 frames and the cue is two seconds, the cue will spill onto the next shot. That spill is the miss. Viewers read a sentence on a pack, a cutaway, or a different speaker.

This is not CPS as a house number. A 40-character line can be finishable in two seconds on a locked camera and unreadable on a four-cut montage of the same line. The characters did not change. The plate did.

Authored jump cuts — you removed the breath on purpose — are a timing job. Artefact jumps inside a generate — a hand teleports, a bottle is suddenly open — are a regenerate job. Do not cover an artefact with a caption. Do not leave an authored jump with a caption that still thinks the breath is there.

Word-highlight makes spill obvious. The accent walks onto a new shot. Static BASIC hides some of it: the whole clause sits on the wrong picture. Both are wrong. Karaoke is not the cause. The un-retimed hold is the cause.

Caption the cut you will ship

Order is load-bearing. Picture lock means the jumps are in.

Wrong loop:

1. Transcribe the long interview or the uncut UGC take.
2. Burn a two-line track timed to that tape.
3. Jump-cut the file for energy.
4. Upload.

The burn is pixels. Trimming a pixel-captioned file does not ripple cues. You cut the picture and keep the old times, or you cut through a glyph. Either way the two-second line is now a stranger on the new shot.

Right loop:

1. Jump-cut, tighten, insert B-roll. Lock.
2. Transcribe this file. 165 language codes. ASR on the cut timeline, not the leftover long take.
3. Split cues at shot changes, not only at clauses.
4. Hold each cue inside the shot it belongs to, unless you mean a line to carry across a cut.
5. Burn.

If you already have a cleaned transcript from the long take, reuse the words. Do not reuse the timestamps. Forced alignment to the cut file is a new clock. Pasting an old .srt onto a jump-cut master is how “I shipped in 12 days” lands on the unboxing B-roll.

B-roll is the usual victim. Editors cut to the product on a pause. The caption from the pause is still up. Mute viewers read the spoken aside on the pack. If the aside was a joke, the pack is now the punchline. If the aside was a price, hope it matches the PDP. Either way you did not place the line. You leftover it.

Carry-across is allowed when the sentence is the point of the montage. Then the hold is a design: one clause, several shots, a finishable duration. Mark it. Accidental carry-across is leftover timing.

Two seconds is a budget on a fast cut

A two-second line is a reasonable default on a locked talking-head. On a cut every eight to twelve frames it is a paragraph.

Fixes, in order:

1. Split at the cut. If the shot is 0.8s, the cue that lives there has to be readable in 0.8s or it should not live there. Shorten the text or drop it.
2. Hold to the shot boundary, then yield. Do not let a cue bleed 400ms into the next face.
3. Skip captions on flash frames. A six-frame cutaway cannot carry a clause. Leave it empty. The eye needs the picture.
4. Stop cutting through a name. Names, SKUs, URLs do not hyphenate and do not split across a jump if you can help it. Hold the shot, or hold the string on one side of the cut.
5. Drop karaoke on dense cut-ups. Snap and DYNAMIC pop need a stable token clock. Fast cuts plus word-highlight is a strobe. BASIC two-line, shorter cues.

All-caps and emoji spend the same budget. A fast cut does not have spare width. Sentence case. No sticker.

Three-line blocks are already a chrome collision on 9:16. Fast cuts make the collision move. Two lines, then split. If the speaker lists four SKUs without a breath, you either cut the list or you accept more cues. You do not “keep up” by leaving one two-second block over four shots.

Safe area still applies. A cue that is correctly timed to a shot can still sit under a username. Timing QC and placement QC are different pauses. Do both on a phone, muted.

Re-time after every recut

A “final” that gets one more energy pass is not final. Each new jump is a recaption.

Practical QC:

• Play the burned file with your eyes on the type, not the face. Write down every cue that is still up after the picture has changed speaker, product, or scene.
• Those timestamps are suspects. If the sentence still belongs, split it to the new boundary. If it does not, kill the leftover.
• Watch once more on mute at phone size. If you cannot finish a line inside its shot, the line is too long for this cut.

The AI caption generator and add captions to a video will time to the file you give them. Give them the cut. 21 BASIC at 1×, 9 DYNAMIC at 2×, 45 Studio CSS presets — none of those is a jump-cut detector.

If you captioned before the edit because a client wanted a preview, throw that burn away. A sample style preview on three seconds is allowed. A full burn of the uncut master is leftover as soon as the first jump lands.

Two-second lines on fast cuts have to be re-timed. The hold follows the shot. Burn after the edit.

Re-time the next cut in Versely.

FAQ

Can I keep one caption track if I only trim the ends?

A trim of pre-roll is an offset. Slip the whole track or recaption. Jumps inside the body still need a re-time. Ends-only is the rare case, and you should still watch the first cue.

Does DYNAMIC karaoke help fast cuts land?

No. Word-highlight needs stable timestamps. Fast cuts expose lag and turn the accent into a strobe. Use shorter BASIC cues, split at the cut.

Should every shot have a caption?

No. Flash cutaways and pack holds often should not. Empty is a choice. A leftover two-second line on the wrong picture is not.