How to caption a short video: pick a style, clear the safe zone

Add subtitles listens to a clip, transcribes what is said, times it word by word, and burns the captions into the picture. What comes back is a finished MP4 you can post.

You choose the look: one of five presets, or your own type, size, colour and grouping. The rest comes from the timing of the speech itself: how long each cue holds and which word is lit.

The tool itself, with the five presets side by side, is on the Add subtitles page. This guide is about the choices around a run: which look suits the piece, where the words should sit, and how to get a transcript worth burning in.

Burned in, not a sidecar

The captions are part of the image. There is no .srt file beside the video and no track a player could switch off.

That is the right answer for the place these go. Social feeds autoplay muted, and a caption track is either unsupported or off by default on most of them, and a subtitle nobody sees is a subtitle that did not exist. Burning them in means the words are there in every player, every embed, every re-upload.

It is also the trade: you cannot turn them off, translate them later, or fix a typo without running it again. For broadcast, for accessibility compliance, or for anything that needs a switchable track, this is the wrong tool and the right one is a proper subtitling workflow outside the studio.

The five presets

The first four come down to two decisions.

Uppercase or not. Pop and Karaoke shout, which is the idiom of a short-form feed and reads at a glance on a small screen. Clean and Boxed do not, which is the idiom of anything that wants to be taken at its word: a product explainer, a testimonial, a piece with a serious subject.

Does a word light up. Pop and Karaoke highlight the word being spoken, which is what makes captions feel synced rather than merely present. Clean and Boxed hold the phrase still, which is calmer and easier to read a long sentence in.

Boxed is the one to reach for when you do not control the background: an opaque box is legible over anything, where an outline or a shadow can still be lost in a busy, bright or high-contrast frame.

One word is the fifth answer: a single big word at a time, timed to the speech. It is the loudest of the five and holds no sentence at all, so it suits a short, punchy line better than an explanation.

Behind the presets is a console. Font, weight, size, case, colour, highlight, outline or plate, and the distance from the top or bottom edge are all controls, so a preset is where you start rather than where you have to stay.

The style is frozen into the result when you run it. Changing your mind is a second run, not an edit.

What it needs from the clip

Three requirements, all checked before any provider work begins, so a clip that fails one of them costs you nothing:

• It must have an audio track. A silent clip has nothing to transcribe.
• It must be decodable video with a real, measurable duration.
• It must be five minutes or shorter. Longer clips are refused outright.

The five-minute ceiling is the shape of the tool rather than an arbitrary limit: these are captions for short-form video, and the cue lengths, the sizing and the pacing are all tuned for a phone-sized frame and a feed-sized attention span.

A run costs the same whether the clip is eight seconds or four minutes.

Getting a clean transcript

Everything the captions get right or wrong starts in the audio.

Clean speech transcribes cleanly. A voice recorded close, a generated voice-over, or dialogue a video model produced are all easy. Heavy background music under the speech, two people over each other, or a distant phone recording in a room are where words start being guessed.

Names, brands and jargon are the weak spot. They are exactly the words a general transcriber has the least reason to expect, and exactly the ones you most need right. Check them before you post; it is the one read-through worth doing every time.

Mixed languages within one clip drift. The transcriber settles into a language and stays there, so a line in another one tends to arrive spelled as though it were the first.

After it runs

The result is an ordinary clip in your library. It lands on the Shorts pill, it gets a poster, it can be shared, and it can itself be a source: a captioned clip can be joined into a longer cut, or have a frame pulled from it.

What it cannot be is edited. There is no transcript editor: the words that come back are the words that were heard.

Fix a wrong caption at the source. Re-record the line, or re-generate the clip with the dialogue written more plainly, and run it again. Fighting the output is not an available move.

When it goes wrong

It refused before it started

One of the three gates: no audio track, a file it could not decode, or over five minutes. Nothing was charged.

Fix: check the clip has sound and is under the cap. If it is a long piece, cut it into parts, caption each, and join them back.

The captions sit over something important

The presets put captions where short-form video puts them, low in the frame and clear of the top. On TikTok, Reels and Shorts that lower band is shared with the app’s own buttons, the account name and the post caption, so low can still mean under the like button.

Fix: before you run it, drop the clip into the free safe-zone checker to see what each app covers. Then move the captions with the position control, top or bottom and how far from the edge, and keep anything else that must be read out of the caption band: logos, prices, faces you have framed at the bottom edge.

A name is spelled wrong

The transcriber heard it correctly and spelled it the common way.

Fix: there is no dictionary to teach and no transcript to correct. If the name has to be right on screen, say it in a clip of its own and title it, or put it in the frame as artwork rather than as speech.