Text people can actually read

Most video on a phone is watched without sound, at least at first. That makes text the primary channel rather than a courtesy, and it makes legibility a structural decision rather than a design flourish.

Legibility beats style, every time

Four things decide whether text can be read on a phone held at arm's length.

Contrast. Light text on a dark plate, or dark on light. Text laid directly over footage fails the moment the footage gets busy, which is why a semi transparent plate behind it is worth the small ugliness.

Weight. Thin typefaces vanish under compression, because the encoder treats a one pixel stroke as noise and smooths it away. Use a medium or bold weight for anything at small size.

Size. If you cannot read it on your own phone from normal viewing distance, nobody can. This is a thirty second test and almost nobody does it.

Time. A viewer needs to read it twice. Four to six seconds for a name and title, longer for a full sentence. Read it aloud twice at a comfortable pace and you have your duration.

Where the interface covers your frame

On the vertical feeds, the bottom fifth of the frame carries the caption text, the account name, the audio credit and a row of buttons. The right sixth carries the like, comment and share column. Anything placed there is covered on the platform and perfectly visible in your editor.

Keep text inside the middle 80 percent vertically and clear of the right edge. If your tool draws a safe area, turn it on and leave it on.

On a television the concern is different and older: the edges of the frame may be cropped. Title safe is roughly the inner 80 percent and action safe roughly the inner 90.

Lower thirds

A lower third is a name and a role. Its whole job is to answer who is this person, and it should do that and leave.

On screen for four to six seconds, once, the first time somebody appears. Bringing it back later says the audience has forgotten, which is occasionally true and usually condescending.

Keep it out of the bottom fifth on vertical, which means a lower third on a phone platform is not actually in the lower third.

Captions are not the same as subtitles

The distinction matters for delivery. A subtitle file sits alongside the video and the viewer can turn it off, restyle it, or have the platform translate it. Burned in text is part of the picture forever.

SRT is the format everything accepts. It is plain text: a number, a timecode range, the lines, a blank line. You can open one in Notepad and fix a typo in ten seconds, which you cannot do with burned in text.

Deliver captions as a file wherever the platform takes one. Burn them in only where it does not, or where the style is part of the piece.

Writing captions that read well

Two lines maximum on screen at once. Around 40 characters per line, which is roughly seven or eight words. Longer lines make the eye travel and the viewer loses the picture.

Break lines at natural phrases rather than wherever the character count ran out. "The county commissioners voted / three to one on Tuesday" reads. "The county commissioners voted three / to one on Tuesday" does not, and the reason is that the brain parses at the break whether the sentence is finished or not.

Keep each caption on screen long enough to read at about 160 to 180 words per minute, which is a little slower than most people speak. Where somebody talks faster than that, condense rather than flashing text nobody can follow.

In AVPrep2. Caption lines wrap at 42 characters, and the wrap happens on word boundaries so a long line becomes two readable ones rather than one that runs off the frame.

Accessibility is not a separate job

Captions serve people who are deaf or hard of hearing, people in a noisy place, people in a quiet place who cannot turn the sound on, and people watching in a second language. That is most of your audience, and the same file serves all of them.

A transcript alongside the video costs nothing extra once the captions exist, makes the content searchable, and is the only version that works for a screen reader.

A checklist

  • Read every title on a phone before you export.
  • Nothing important in the bottom fifth or the right sixth on vertical.
  • Names on for four to six seconds, once.
  • Captions two lines, about 40 characters, broken at phrases.
  • SRT where the platform takes it, burned in where it does not.
  • Spell the names right. It is the one error the subject will always notice.