Word by word captions: how karaoke-style captions are made

What word by word captions are, and why short-form uses them

You've seen the look on TikTok, Reels and Shorts: three or four words on screen, big and bold, and the word being said right now turns yellow or sits in a colored box. The next word lights up as the speaker gets to it. The style goes by a lot of names (karaoke subtitles, highlighted captions, animated captions), but it's one technique.

Creators use it for plain reasons. Plenty of people scroll with the sound off, so the words have to carry the video. A short line is easier to read on a phone than a full sentence. And the moving highlight tells the viewer where to look, which keeps the eye on the text instead of drifting away. It isn't magic for retention, and a boring clip with fancy captions is still boring. But on a talking-head video it's the difference between readable and skippable.

For long YouTube videos, films and courses we'd skip it. A highlight bouncing along for 20 minutes is tiring, and viewers there expect captions they can turn off.

How karaoke captions work

Three steps, none of them complicated.

1. Word-level timestamps

Normal subtitles only need a start and end time for each line. Karaoke captions need a start time for every single word. Our auto captions tool asks the speech model (Whisper large-v3 turbo) for word timestamps, so a sentence comes back something like this:

2. Short lines

Next the words are grouped into lines. For a vertical video our tool keeps each line to about 22 characters and no more than 2.5 seconds on screen, so you usually see two to four words at a time. It also breaks at pauses and at the end of a sentence, so a line doesn't straddle two thoughts. Landscape videos get longer lines of up to about 42 characters.

3. The highlight

For every frame, the renderer finds the line that's on screen, then walks through its words and picks the last one whose start time has already passed. That word gets the highlight. Everything else in the line stays white. At 1.0 seconds in the example above, "one" has started and "second" hasn't, so "one" is lit.

One detail you'll notice: the highlight moves on when the next word starts, not when the current one ends. During a short pause the last word stays lit, which looks calmer than a highlight that blinks off and on.

The same drawing code runs the live preview and the final export, so what you see while editing is what ends up in the MP4.

Design rules for highlighted captions

• Two lines at most, and on vertical video three to five words per line. If a line needs more, split it.
• Keep it clear of the app's interface. TikTok and Reels put the username, post text and buttons along the bottom and the right edge. Captions at the very bottom of the frame end up under them.
• Use high contrast. White text with a thick dark outline, or text on a solid box, stays readable over any footage. Thin text with no outline disappears on a bright background.
• Check the highlight color against the footage, not just against white. Yellow on a dark outline is easy to read. Red or dark blue on dark video isn't. If you use a color change, the highlighted word still has to be readable on its own.
• Don't rely on hue alone. Some viewers can't tell red from green, so a red-to-green switch may look like nothing happened. A box behind the word changes shape as well as color.
• Fix names and numbers before styling. A highlighted typo is a very visible typo.

How to make word by word captions for free

1. Open the auto caption generator and drop in your video (MP4, MOV or WebM, up to 30 minutes). The video stays in your browser. Only a compressed audio track is sent for transcription.
2. Wait for the transcript. Vertical videos switch to short lines automatically. For a landscape clip you want in short chunks, tick Short lines for Reels & TikTok in the Style tab. Do this before you edit, because it re-splits the captions and replaces text changes.
3. Read every line. Click one to fix a word, and nudge its start or end time if it's off.
4. Pick a preset in the Style tab: Karaoke or Bold pop (see the table below).
5. To change the highlight color, use the Highlight color picker. The Active word menu switches between Highlight box, Color change and No highlight.
6. Drag the Position slider up if the captions sit under the app buttons. Karaoke starts at 86% of the frame height, which is inside the area TikTok and Reels cover, so around 70% is safer.
7. Open the Export tab and click Render video. Your browser encodes the MP4 (or a WebM in browsers that can't write MP4), with no watermark.

One tip from building this: when you fix a typo, keep the same number of words and the timing stays exact. If you add or remove words, the editor spreads that line's timing across the new words by length, which is close but not perfect. The same happens with an imported SRT, since it has no word times to begin with. For the platform side of posting, see how to add captions on TikTok and how to add captions to Instagram Reels.

If you already edit in CapCut or Premiere Pro

Stay there. Moving to another tool for one caption style isn't worth it if the rest of your edit lives in one app.

In CapCut, generate auto captions first, then pick a caption template. Several of them highlight the spoken word, and many add motion like bouncing or scaling, which our tool doesn't do. Menu names shift between the phone app, desktop and web versions, so our CapCut auto captions guide walks through each one.

Premiere Pro is further behind here. Version 26.3 (June 2026) added Single Word captions, which puts one word on screen at a time, synced to the audio. Adobe's help pages don't describe a built-in way to highlight the spoken word inside a full line, so editors fake it with keyframes or use a plugin. Or make the captions here, export the MP4, and drop it on your Premiere timeline. Our guide to captions in Premiere Pro covers the standard caption workflow.