ThreeCutBlog

← All posts

How to add word-by-word captions to your videos

· 6 min read

A paper cut-out phone showing a smiling man from the chest up, with three blank paper word tiles below it and the middle one coral and raised

Word-by-word captions, also called karaoke captions, show a short phrase on screen and highlight each word at the moment it is spoken. To add them, transcribe the audio with a timestamp for every word, correct the transcript, group the words into chunks of two to four, and burn them into the video with the current word highlighted. Size and placement matter as much as timing: on a vertical video they need to be large, bold and clear of the buttons the app puts over your picture.

They work because the eye follows the highlight at the speed of the voice. Someone watching with the sound off can still follow every sentence, and someone with the sound on reads along instead of drifting.

The rules that make them readable

Setting What works Why
Words per chunk 2 to 4, about 22 characters at most A full sentence on screen gets read ahead of the voice
Lines 1, or 2 at most More lines push into the picture and the platform's buttons
Highlight The spoken word in a contrasting box or colour Pulls the eye along at speaking pace
Words already said Full white Easy to reread
Words still to come Dimmed to about half Shows what's coming without competing
Font Bold sans-serif, white, dark outline or shadow Readable over any background
Punctuation Drop full stops at the end of a chunk; keep commas, question marks and exclamation marks Full stops add clutter; commas keep "New York, London" from reading as one place

Two more rules that are easy to miss:

  • Start a new chunk at the end of a sentence and after a pause. A chunk that straddles two sentences makes the viewer read half of the next thought before you've said it.
  • Never split a name across chunks. "New" on one screen and "York" on the next reads as two words. Keep capitalised pairs together.

How big to make them

On a phone held at arm's length, captions need to be much larger than they look on a laptop while you edit. A good starting point is a font size of about 7 to 10 percent of the frame's width. On a 1080 by 1920 vertical video that is roughly 75 to 105 pixels.

Check it the honest way: export a 10 second test, open it on your phone, hold the phone where you normally hold it, and read it without squinting. If you have to lean in, go bigger.

Longer words will not fit at full size inside the safe area. Shrink that chunk rather than letting it run off the edge, and keep the size the same for the rest of the video so the captions don't pulse.

Where they go: the safe area

Every app puts something over your video. On Reels, TikTok and Shorts the bottom of the frame holds the caption and account name, the right edge holds the like, comment and share buttons, and the top holds the app's own bar. Text placed there gets covered.

These are the margins ThreeCut keeps text out of, as a share of the frame. They are a reasonable default for any editor:

Shape Top Right Bottom Left
9:16 vertical 8% 14% 20% 6%
1:1 square 6% 6% 10% 6%
16:9 landscape 6% 6% 9% 6%

In a vertical video, that puts the last line of the captions roughly three quarters of the way down the frame: below the speaker's face, above the account name. Centre the captions on the safe area, not on the frame, because the right margin is wider than the left.

If you also use on-screen headlines, put them near the top of the safe area so the two never stack on each other.

How to time them

Word-by-word captions only work if the timing comes from the real audio, not from an estimate. A highlight that runs ahead of the voice is worse than no highlight.

  1. Transcribe with word-level timestamps. Most transcription tools can return the start and end time of every word.
  2. Correct the transcript before you style anything. Product names, people's names and numbers are where transcription slips. A caption that spells your product wrong is on screen for everyone to see.
  3. Show each chunk just before its first word. A few hundredths of a second early (around 40 milliseconds) feels in sync; exactly on time can feel late.
  4. Move the highlight on each word's start time. Hold it on a word until the next word begins.
  5. Hold the last chunk briefly at a pause. Let it stay up for about half a second after the last word, then clear it, so the screen isn't empty mid-sentence but doesn't hang on old words either.

If you cut the video after captioning, the timings are wrong. Edit first: remove the ums, false starts and long pauses (see how to remove filler words), then caption the edited audio.

Burned in or uploaded?

You can burn captions into the picture, or upload a caption file (such as SRT) that the platform displays. For short social video, burn them in: platforms show uploaded captions in their own small style, without the highlight, and not every viewer turns them on. Where a platform accepts a caption file as well, such as YouTube, uploading one too helps people who need captions in their own size or format.

Common mistakes

  • Too many words at once. Six words in two lines turn into subtitles, and the highlight has nothing to do.
  • Captions under the account name. They look fine in the editor and are half covered in the app.
  • A highlight colour that disappears. A yellow highlight on a sunny beach shot vanishes. Put the spoken word in a solid box, or check every scene.
  • Captions on top of the face. The middle of the frame belongs to the speaker's eyes and mouth. Keep text below them.
  • A wrong word. One misheard word undoes the trust the rest of the video built. Read the captions once, slowly, before you post.

For the rest of the edit, from the first word to the last, see how to edit a talking head video.

How to do this in ThreeCut

Every ThreeCut video gets word-by-word captions without any setup. After you record, or after the AI narrator reads a faceless script, ThreeCut transcribes the audio and times each word to the real voice. Captions appear in chunks of up to four words, starting a new chunk at the end of a sentence or after a pause. The spoken word sits in a coral box, words already said are white, and words still to come are dimmed. Text is placed inside the safe area of the shape you picked (9:16, 1:1 or 16:9), and because the captions are made from the edited take, the ums and false starts the edit removed never appear on screen. See how it works.

Common questions

What are karaoke captions?

Karaoke captions are word-by-word captions: a short phrase stays on screen and each word is highlighted as it is spoken, like lyrics on a karaoke screen. They are common on Reels, TikTok and Shorts because they keep the viewer's eye moving with the voice.

What size should captions be on a 9:16 video?

Start with a bold font at about 7 to 10 percent of the frame's width, which is roughly 75 to 105 pixels on a 1080 by 1920 video. Then check it on a phone at arm's length and go bigger if you have to lean in.

Where should captions go on Reels and TikTok?

Below the speaker's face and above the bottom fifth of the frame, centred within the safe area. Keep them away from the right edge, where the like, comment and share buttons sit, and from the bottom, where the app shows the caption and account name.

How many words should be on screen at once?

Two to four words, and no more than about 22 characters. Start a new group at the end of each sentence and after each pause, so the captions follow the rhythm of speech.

Read next