Word-by-word captions,and how YouTube plays them

Captions that arrive one word at a time are everywhere in short-form video, and they work: the viewer can't read ahead, the eye stays near the centre of the frame, and fast speech becomes easier to follow. YouTube can play them back from a caption track, without burning anything into the picture — but only in a particular way.

Today we fix that.

Appear Highlight Whole line

Appear and highlight are not the same feature

Appear: words are added as they are spoken

The line starts empty and each word shows up at its own moment. In YouTube's caption format this is native: one caption line contains several segments, and each segment carries its own start offset. The player does the rest, the file stays small, and the timing is exact.

Highlight: the whole line is visible, the current word is coloured

Here the viewer reads the full line from the start while one word lights up. The format has no "recolour this word now" instruction, so the effect is built by writing the same line many times — one copy per word, each with a different word coloured. A ten-word line becomes ten caption entries.

That is why highlight is the heavier choice, and why a centred line built badly jumps around: if each copy is rendered as a fresh line, the player re-centres it. Done properly, every copy contains identical text with identical spacing, and only the colour changes.

Roll-up and zoom

Two more effects show up in the same conversation:

• Roll-up — lines stack and scroll upwards like live captions on television. Each state of the stack is its own caption entry.
• Zoom — the line grows into place. YouTube has no animation instruction either, so the movement is written out frame by frame: a series of entries a few dozen milliseconds apart, each with a slightly bigger font size.

Both are legitimate and both inflate the file. A three-minute video with word-by-word appear captions stays a modest XML file; the same video with zoom on every line can be tens of times larger.

Timing: the part that decides whether it feels right

Word timing usually comes from one of two places. Speech recognition gives a timestamp per word, which is as accurate as the transcription. Splitting a line's duration between its words is the simpler route: divide by word count, or weight each word by its length so that "extraordinary" holds the screen longer than "and".

Proportional timing is good enough for most talking-head footage and noticeably off when someone pauses mid-sentence. Rules that hold either way:

• Two to five words per line; a long line defeats the effect.
• Around 250–400 ms per word feels natural; below 150 ms words flicker.
• Break lines where the speaker breathes, not where the character count runs out.
• Don't start the first caption at 0 ms — some Android players ignore a line that starts at zero. Start at 1 ms or later.

What survives translation

Word-by-word timing is tied to the words of one language. Translate the line and the word count changes: four English words can become six in German or two in Japanese. Timings have to be redistributed across the new words, and any per-word colouring has to be re-applied, which is exactly the kind of work that makes hand-translating styled captions miserable. Rainbow Captions redistributes the timing for every translated language automatically and keeps the styles.

Where these captions actually play

Word-by-word captions from a caption track play in YouTube's web player and in the mobile apps. Support is not uniform across every device and TV app, and viewers who have set their own caption preferences in the player will see their settings win over yours. Lines that mix several styled segments also need invisible zero-width characters as padding, or some mobile players collapse the styling — a quirk we describe in the .ytt format reference.

Questions

Can YouTube show word-by-word captions without burning them into the video?

Yes. YouTube's caption format supports segments with their own start times inside a caption line, which is exactly the appear effect. Highlight, roll-up and zoom are possible too, but they are written out as repeated caption entries rather than as animations.

Which is better, appear or highlight?

Appear is lighter, more precise and the closest to what large channels use. Highlight suits lyrics or teaching material, where the viewer benefits from seeing the whole line while following the current word.

Do animated captions hurt accessibility?

They can. Viewers who read slowly, or who rely on captions rather than audio, lose the context of the full sentence when words arrive one at a time. Keep lines short, keep the pace steady, and publish a plain caption track alongside the styled one when the audience needs it.