Caption Timing Rules That Make Subtitles Readable

Bad subtitles are almost never bad because the words are wrong. They are bad because of when they appear, how long they stay, and where they break — and the viewer experiences all three as a single vague feeling that the video is hard to follow. The symptom people describe is "the subtitles are distracting". The cause is usually a reading-speed figure nobody measured.

There is a small set of numbers that governs this, and they are stable across languages and formats because they describe the eye rather than the file. This guide covers what those numbers are, why each one has the value it does, how to break lines so the reader does not stall, and what to do when a file you have already been given runs too fast to read.

The Four Numbers

Almost every timing problem reduces to one of four measurements. Learn to recognise which one you are looking at, because the fixes are different and applying the wrong one makes the cue worse.

These figures come from long-standing subtitling practice rather than from any single authority, and different style guides place the boundaries a character or two apart. Treat them as the conventions they are: the range where most viewers are comfortable, not a threshold where comprehension collapses.

Reading Speed Is a Budget

Characters per second is the measurement people skip, because it requires arithmetic and the other three can be eyeballed. It is also the one that decides whether the subtitles work.

Divide the character count of the cue by its duration in seconds. A 68-character cue on screen for four seconds is 17 CPS — right at the ceiling, and the viewer will spend nearly the whole four seconds reading. The same 68 characters over five and a half seconds is about 12 CPS, which leaves room to look at the picture, which is the point of watching a video.

That framing matters more than the number itself. You are not trying to hit 15 CPS; you are trying to leave the viewer enough spare attention to see what is happening on screen. Every character you cut buys back time. This is why professional subtitling condenses rather than transcribes — "I think what we should probably do here is" becomes "We should", and nothing is lost except reading load.

Condensing text the viewer can still hear needs a light touch, though. Drop a word that is clearly audible and the mismatch registers, even if the reader could not say what was wrong. The comfortable version is to cut what carries no meaning — false starts, repeated subjects, hedges — and keep everything that does. Where those stumbles are also being cut from the audio, the two jobs should be done together so the text follows the finished edit rather than the raw take, which is the approach in removing filler words without making speech sound chopped.

Two situations legitimately push the budget. Rapid overlapping dialogue leaves no gap to expand into, so condensation is the only lever you have. And a cue containing a number, a name or a technical term needs more time than its character count suggests, because unfamiliar strings cannot be read at speed.

Break Lines Where the Grammar Breaks

Line breaks are the part of subtitling that looks like formatting and behaves like editing. The eye reads a line, jumps down, and reads the next. If the first line ends mid-phrase, the reader has to hold an incomplete unit across that jump, which costs time you already budgeted away.

The rule is to break at the largest available grammatical boundary. Between clauses is best. After a complete phrase is fine. Splitting an article from its noun, a preposition from its object, or a first name from a surname is not.

A concrete pair makes the difference obvious:

Bad:
We moved the interview to the
second location because of noise.

Better:
We moved the interview to the second location
because of noise.

The second version breaks between clauses, so each line is a readable unit on its own. The first strands "the" at the end of a line, where it means nothing until the eye has travelled.

The same logic governs the break between one cue and the next, and there the cost is higher. A line break asks the eye to travel; a cue break removes the text entirely and replaces it. Ending a cue on a conjunction leaves the viewer holding a sentence fragment through a blank moment, and when the continuation arrives they have to reconstruct the first half from memory before they can read the second. Split between sentences wherever the timing allows it, and where it does not, split after a clause so both halves stand alone.

Where two lines are unbalanced, prefer the shorter line on top. A long first line followed by a short second one draws the eye down and to the left across a big empty gap. The reverse shape is easier to track and takes up the same space.

Shot Changes Are Harder Than Sentences

A cut is a visual event, and the eye responds to it by re-reading whatever text is on screen. That reflex is why a cue that crosses a cut by two or three frames feels broken even though nothing is technically wrong: the viewer re-reads the same sentence they already finished.

The convention is to snap. If a cue ends within a few frames of a cut, move its out-point to land exactly on the cut. If it starts within a few frames after one, pull its in-point back to the cut. Cues that genuinely need to span a cut should span it by a comfortable margin — a second or more on either side — so the text is clearly continuous rather than accidentally overlapping.

This is one place where caption work and edit decisions are entangled. If you find yourself fighting a cut every few seconds, the cutting rhythm and the dialogue rhythm are out of phase, which is a problem in the edit rather than in the subtitle file. The relationship between cut placement and what the viewer is doing at that moment is covered in choosing where a cut lands.

Where Short-Form Legitimately Departs

One-to-three-word kinetic captions do not obey the reading-speed budget, and it is not sloppiness — they are doing a different job. Those cues are locked to the spoken word, so their duration is set by speech rather than by reading, and characters-per-second stops being a meaningful measurement.

What still applies is the floor. A word that appears for four frames is a flicker regardless of style, so words spoken very quickly should be grouped rather than split. Placement still matters too, because a cue positioned where the platform's interface sits has not been delivered at all.

The trade-off between the two approaches, and the cases where word-level captions actively hurt comprehension, is the subject of word-level and line-level captions.

Repairing a File That Runs Too Fast

Fixing timing on an existing file is a mechanical process, and doing the steps out of order creates work. Condense before you retime, because condensation changes the duration each cue needs.

Next Steps

Export the subtitle file from your last finished project and sort the cues by reading speed. If more than a handful sit above 17 CPS, the problem is that the text was transcribed rather than condensed, and the fix is editorial rather than technical.

Then check what your delivery format can actually carry, because a cue you have carefully placed at the top of the frame will only stay there in some formats — the differences are set out in the three subtitle formats and what each one drops. Once timing is settled, legibility is the remaining variable, and it depends almost entirely on how the text is separated from the footage behind it. The wider set of caption decisions sits under captions and subtitles.