How to Subtitle Video Ads for Sound-Off Viewing

Start from a word-accurate transcript of the finished audio, not from the script doc you meant to shoot. Creators and avatars drop words, swap numbers, and ad-lib offers. Then strip ums, false starts, and repeated hedges so the on-screen line is readable. The captions should match the voice closely enough that a hearing viewer is not fighting the text, and cleanly enough that a mute viewer can still parse it.

Play the video once with a text file open and correct every number, product name, and CTA verb. Those are the words that get you policy problems and refund requests when they are wrong. If the speaker said a different price than the landing page, fix the video or the page — do not 'clean up' the caption to the page price while the audio says something else. Mute viewers who later unmute will notice, and so will reviewers. A caption that 'fixes' a spoken error is a second, conflicting claim. Flag every digit in the transcript in bold in your notes so a skim cannot miss a price error.

Minor trims are allowed: drop 'like,' 'you know,' and a doubled 'really.' Do not rewrite the line into marketing copy the mouth never said. If a sentence is too long to ever read, the fix is a tighter read or a cut, not a caption that paraphrases a different ad. Keep disclaimers that were spoken; hiding them in the transcript while they remain on the audio is how you get a mismatch, and dropping them from both still may not be legal if the claim needs them. When in doubt, keep the spoken legal line on screen in a short block of its own.

Split the transcript into caption blocks of about four to six words, one idea each, in and out with the spoken phrase. Do not park a whole sentence on screen for four seconds. Mute viewers read in pulses; a wall of text is the subtitle version of a monotone. Break at clauses, not in the middle of a product name or a number, and not after a hanging 'the' that means nothing alone.

A good split follows breath. 'This cream changed my skin' then 'in a week' is readable; 'This cream changed' then 'my skin in a week' is a hitch. When two short clauses belong together, keep them in one block rather than flashing three words that mean nothing alone. Time each block to appear within a couple of frames of the spoken start and to leave before the next idea, with a brief gap so the eye notices the change. If two blocks overlap, you have not timed them — you have stacked them. If a block is still on screen after the speaker has moved on, you are late — pull the out-point forward.

If a fast talker piles three claims into two seconds, you cannot subtitle all of them. Keep the claim that the picture also shows, and cut the extra talk — captions cannot rescue a crowded read. For numbers, never split a figure across blocks ('$49' then '99'). For names, never split a SKU. Watch the timeline while you listen once; timing that only looks right as a subtitle file will drift the moment speech has a pause you did not mark. Burn the timed file against the actual take, not against the script's estimated duration.

In each block, mark the number, the benefit, or the CTA verb for emphasis — not the whole line. Keyword highlighting is how a mute scroller gets the point in one glance. If every word is bold and bright, you have not highlighted anything. Match this pass to meaning, then let styling follow, rather than decorating first and hoping the right word pops on a busy B-roll cut.

Prefer highlighting what you would still want understood if the rest of the line vanished: '3 days,' 'no drip,' 'shop the bundle,' 'book the slot.' Skip highlighting filler verbs and brand slogans unless the slogan is the hook. Keep the same highlight rule across the video so the eye learns the pattern. If you localize later, highlight the equivalent meaning in the new language, not the same English word sitting awkwardly in a translated sentence. One highlight color, used sparingly, beats a rainbow of emphasis that reads as noise.

Do not let emphasis cover the product or fight a busy B-roll cut. If the demo beat is the proof, you can drop to a lighter caption or fewer words rather than stacking a glowing keyword on top of the thing you want people to see. Emphasis is a reading aid for sound-off, not a sticker pack. Recheck the hook block especially; if the emphasized word is not the reason to stop, pick a different word or a different hook. A mute skim of highlights only should still sound like an ad, not like a random noun list.

When the ad will run in a second language, rewrite the subtitle file for that market: line length, idiom, offer wording, and units. A machine dump of the English transcript will overflow the frame, miss the joke, and sometimes invent a claim. Localization is part of the sound-off job because mute viewers in that market never hear your original line to correct it, so the text is the whole ad.

Have a native speaker time the new lines against the same video. Some languages expand and will not fit a four-word English block; split or recut rather than shrinking type into unreadable captions. Keep product names as you must legally keep them, but translate the CTA into the verb people actually tap in that market. Re-highlight meaning in the new language. Recheck numbers, dates, and currency so the caption does not disagree with the on-screen price or the landing page. A localized subtitle that still shows an English offer code is a half-done file.

Do not auto-translate disclaimers into something stronger or weaker than counsel approved. If a market needs extra legal copy, you may need a dedicated caption block or an end card, not a microscopic line. If the avatar's mouth is clearly English and the captions are another language, that can still work for sound-off, but then the captions have to carry 100% of the message — which is a reason to be stricter about timing and accuracy, not looser. Never ship a translated file you have not watched muted in that language.

Export and watch the subtitled video on a physical phone, muted, at arm's length, in the apps you will run — or in their ad previews. Confirm every line is readable, no block sits under TikTok's caption bar or a Shop button, no keyword is covered by the product, and a mute stranger can retell the offer after one play. Desktop players lie about size and about overlays, every time.

Klip Kanvas auto-generates captions you should still treat as a first pass: trim line length, fix names, retiming to the take you kept, and re-highlight the claim words before you export. Then do the phone check on that export, not on the editor canvas. If a line is late, pull it forward; if two blocks collide, you overcut. Check the first two seconds twice. A mute hook with no readable text is a skipped ad, even if the audio hook is strong. If the first block is a filler word, you timed from the wrong start.

Keep a short QA list: transcript matches audio on numbers, blocks are one idea, timing tracks speech, highlights are sparse, safe from UI, localized file is a real rewrite. Fail the video if a mute viewer cannot name the product and the next step. Styling choices — font, outline, box — belong in your caption style, but they will not save a late, long, or wrong subtitle. Fix the words and the timing first, then restyle. A beautiful caption that says the wrong price is still a wrong ad. Fail the pass if a teammate watching muted cannot say the CTA verb back to you in one try.