Captions people actually read.
A lot of people meet your video with the sound off. How to caption so the words land on the beat and stay out of the app’s way.
Markers (6)
A lot of people meet your video with the sound off. On a train, in a queue, in bed. For them the captions are the script.
Most captions are made as if nobody will read them: one long line, dropped in late, sitting where the app puts its buttons. Here is how we caption short form so the words land when they are said and stay where a thumb will not cover them.
Three to five words a block
A caption is read in a glance, or not at all. The eye takes in a few words at once, so give it a few words at once. Three to five is the range we work in. Fewer and the screen flickers. More and the viewer is reading instead of watching.
A caption is read in a glance, or not at all.
Break the lines where a speaker would breathe, not where the box runs out. “the one setting” and “that ruins it” read as two thoughts. “the one setting that” and “ruins it” read as a riddle. Keep the small words with the words they belong to.
Time them to the voice, then to the beat
Automatic captions get the timing close. Close is the problem. A caption that arrives half a second after the word feels like a bad dub, and a caption that leaves early makes the viewer chase it.
So we time in two passes. The first pass is the voice: each block comes on with its first word and goes off as the last word ends. Nudge the edges with the waveform open until the text and the mouth agree.
The second pass is the edit. Where a block ends a frame or two before a cut, pull it to the cut. The words should change when the picture changes, so the viewer’s eye resets once, not twice.
- No block shorter than the time it takes to read it once, even if the speaker is fast.
- When the speaker does stop, let the screen go clear. Silence can be captioned with nothing.
Two lines at most, one highlight colour
Two lines is the ceiling. A third line turns a caption into a paragraph. If a block needs three lines, it needs to be two blocks.
Style is where most captions shout. Pick one typeface, heavy enough to read over a busy frame, with a soft shadow or a quiet box behind it so it survives a white wall or a bright sky. Then pick one highlight colour and use it for one word per block, at most: the word the speaker leans on.
Highlighting every noun is the same as highlighting nothing, only louder.
Stay clear of the buttons
In the editor your vertical frame is clean from edge to edge. On a phone it is not. Down the right side sits a column of buttons. Along the bottom sit the account name and the post’s own caption. Put your words there and the app puts its words on top of them.
The fix is a habit, not a trick. Keep captions in the middle band of the frame: above the bottom strip, left of the button column, and a little off centre towards the top of that band so they sit near the speaker’s face, where the viewer is already looking.
Every app draws this furniture a little differently, and redesigns it without asking. So we keep a generous margin on the right, a deep one at the bottom, and check the export on a real phone.
Try it below. Move the caption down into the bottom strip, or switch to one long line and watch where its end goes.
Live demo·captions_final_v3
Where the caption lives
Show the app furniture, then drag the caption or use the Position slider. Switch to one long line and watch where its end goes.
the one setting that ruins it
Position540.0 1114.0
Show app furniture
Clear of the buttons.
Burned in or a file?
There are two kinds of captions, and they do different jobs. Burned in captions are part of the picture. You choose the font, the colour and the frame each block appears on, and they look the same everywhere. They cannot be switched off or fixed without a new export.
A caption file, usually an SRT, travels next to the video. The viewer can turn it off or on, and on platforms that support it they can change its size in their own settings. It can be corrected without touching the edit, and translated as separate files.
Our rule of thumb:
- Short form for feeds: burned in, styled, timed to the cut. The look is part of the video.
- Long form on a platform that takes caption files: upload a file, and burn in only the lines that matter to the story, like a key quote or a name.
- When in doubt, deliver both. A file from captions you have already timed is quick to make, and it is the version you can still change next week.
Proofread on a phone
Captions are the only part of the edit that can be plainly wrong. A cut can be a matter of taste. A misspelled name is just misspelled, in large bold letters, over your face.
Proofread the export, not the timeline: the timeline shows the text you meant, the export shows the text you made. Do it on a phone.
- Read every block with the sound off. Does each one make sense on its own, without the voice?
- Watch it with the sound on. Does each block land with the word, and leave when the word does?
- Check names, numbers and brand spellings against the brief, one at a time.
- Look at the edges. Is anything under the buttons, or cut off by a rounded corner?
If the video reads well muted, the captions are done. If it only makes sense with sound, better you find out than your audience.