How to add subtitles to a video

Adding subtitles takes four steps: turn the speech into text with the time of each word, correct that text, decide how it should look, and either burn it into the picture or save it as a separate file. Automatic transcription does the first step in minutes. The quality of the result is decided in the second and third.

Most of the work is not typing. A speech model writes the draft; what makes subtitles good is what you do after: fixing the words it got wrong, keeping each line short enough to read, and putting the text where nothing covers it.

First decide: burned in, or a separate file

There are two kinds of subtitles, and they are not interchangeable. Burned-in subtitles are part of the picture. They look exactly as you designed them on every platform and every device, and nobody can switch them off. A subtitle file (.srt or .vtt) is a list of lines with their times, uploaded next to the video. The viewer can turn it on or off, and the platform decides how it looks.

As a rule: burn them in for short clips meant for feeds that play without sound, where the subtitle is part of the look. Use a file for long videos on platforms that support one, such as YouTube, where viewers expect to choose. Nothing stops you from doing both from the same text.

How automatic transcription works

A speech model listens to the audio in short windows and writes the words it hears, each with the moment it starts and ends. Those per-word times are what make the animated styles possible: a word can light up exactly when it is spoken. The words are then grouped into lines, breaking at pauses and at the end of sentences.

Where the model runs matters. A cloud service receives your audio, transcribes it on its own machines and is fast. A model running in your browser keeps the file on your computer, costs nothing per use, and pays for that in two ways: it has to be downloaded once on each device, and it is slower, more so on a machine without a graphics card the browser can use.

Correct the text before anything else

An automatic transcript is a draft, and the errors are predictable. Check these first:

• Names and brands. The model writes what a name sounds like, not how it is spelled.
• Numbers and units. Dates, prices and measurements are often written out or misheard.
• Words that sound alike. Their and there, to and too: correct in sound, wrong on screen.
• The start and end of the video. Music, an intro sting or silence can produce words that were never said.
• Overlapping voices. When two people talk at once, one of them is usually lost.

Read the whole thing once with the sound on. It is faster than it sounds, because you are only confirming, and it is the single step that separates subtitles people trust from subtitles people notice.

Keep each subtitle short enough to read

A subtitle is read while the picture keeps moving, so it has to be taken in at a glance. Broadcast and streaming guidelines commonly keep a line to roughly 37 to 42 characters and a subtitle to two lines. On a vertical clip with large text, two to four words at a time is normal.

• Break where the sentence breaks. A line that ends on "the" or "and" makes the reader wait for the next one.
• Give short lines enough time. A subtitle that flashes for a fraction of a second is not read, however short it is.
• Do not let text outlast the speech by much. A line still on screen when the next person talks is confusing.
• Match the pace of the video. Fast cuts want short subtitles; a calm explanation can carry full lines.

Make it readable on any background

The picture behind the text changes every second, so contrast cannot depend on it. There are two reliable ways to get it: a dark outline around light text, or a box behind the text. An outline is lighter and suits large, heavy type. A box is the safest choice over busy footage and for smaller text.

• Use a heavy, plain typeface. Thin and decorative fonts break up at small sizes and on compressed video.
• Stay with white or a light yellow for the text unless the design has a reason not to. Coloured text needs a stronger outline.
• Keep one colour for emphasis. A highlighted word works because the others are not highlighted.
• Check it on a phone. Text that looks generous on a monitor is often too small at arm's length.

Put it where nothing covers it

On a horizontal video the convention is the lower part of the frame, centred, with a margin from the edge. On a vertical video for a short-video app that position is the worst one: the caption, the buttons and the progress bar sit on top of it. Move the subtitles up, a little below the middle, and keep them away from the right edge where the icons are.

If the video already has text on screen, such as a lower-third name or a slide, move the subtitles for the whole video rather than letting the two collide.

Motion: a little goes a long way

Animated subtitles hold attention because the eye follows what moves. The useful kind follows the speech: a word appears or lights up as it is said. Motion that is not tied to the speech, such as text bouncing for its own sake, competes with the picture. For long videos, a quiet fade or no animation at all is easier to watch for ten minutes than a pop on every word.

Doing it in CutNoodle

The subtitle tool does all of the above in a browser tab. It transcribes on your computer, so the video is not uploaded; the first time on each device it downloads the speech model once, between about 40 and 560 MB depending on the model you choose and on your browser, and keeps it for the next videos. You then correct the lines, set how many words each one holds, pick a style and an animation, drag the text where you want it, and export an MP4 with the subtitles burned in. The same text can be downloaded as .srt or .vtt for platforms that take a file.

It has limits worth knowing before you start: it does not translate and you cannot choose the language, one style applies to the whole video, and the video can be up to 60 minutes and 2 GB.

Common questions

CutNoodle's subtitle tool is free and runs in your browser. Open it and choose a video.