How to make subtitles from scratch (timing & reading speed)

People assume subtitling is transcription — type what you hear, done. Transcription is only the first third of the job. A subtitle that carries the right words but flashes past too fast, or sits on a two-line block you cannot finish reading, is a failed subtitle even though every word is correct. The craft is in the timing, and timing follows a small set of numbers you can learn in an afternoon.

This guide walks the whole path from a blank file to a clean subtitle track: how to transcribe without over-writing, how to anchor each line to the audio, and the three limits — reading speed, line length, and duration — that separate a comfortable read from a stressful one.

What are the three stages of subtitling?

Every subtitle file is built in the same order, and doing them out of order is what makes the work painful. Separate them deliberately.

1. 1Transcribe — get the words down as plain text, one thought per line, ignoring timing entirely.
2. 2Time — attach a start and end to each line so it matches when the words are actually spoken.
3. 3Tidy — adjust the timing so nothing reads too fast, no line is too long, and no cue is too brief.

The reason to keep transcription and timing apart is focus. When you try to type and time at once you do both badly — you either mistime because you are hunting for the next word, or you mistype because you are watching the clock. Write the whole transcript first, then time it in a second pass. The text-to-subtitle tool is built around exactly this split: you paste finished text and it lays out the cue structure, leaving you free to refine the timing.

How do you transcribe cleanly?

A subtitle transcript is not a verbatim court record. Your job is to convey what was said in a way a viewer can read in the time available, which usually means light editing: dropping filler like "um" and "you know", merging false starts, and breaking long sentences at natural pauses. Each line should be one complete idea, because that is what will become one cue.

• Break at natural pauses — the end of a clause or sentence, not mid-phrase. A viewer reads a whole thought more easily than a fragment.
• Cut filler words unless they carry meaning. "Um" adds nothing but characters the viewer must process.
• Keep one speaker per line, and mark speaker changes with a dash so a viewer following a conversation never loses the thread.

How do you time a subtitle to the audio?

Timing means giving each line an in-point (when it appears) and an out-point (when it disappears). The in-point should land as the speaker begins the phrase — a fraction early is fine, late is jarring. The out-point should hold until the viewer has had time to read, which is often a beat after the speaker has stopped, not the instant they finish.

The common beginner mistake is snapping the out-point exactly to the end of the audio. Speech ends faster than reading, so a cue that vanishes the moment the mouth stops leaves the slower half of your audience mid-sentence. Let the out-point breathe. When two lines butt up against each other, leave a small gap between them so the eye registers that a new subtitle has arrived rather than seeing one long block mutate.

Once the rough timing is in, load the file into the subtitle editor to nudge individual cues against a waveform, or check the whole track at once with the subtitle timing checker, which flags every cue that breaks a reading-speed or duration rule so you are not eyeballing hundreds of lines by hand.

What reading speed is comfortable?

Reading speed is the single most important number in subtitling, and it is measured in characters per second (CPS): the number of visible characters in a cue divided by how many seconds it is on screen. A cue with 60 characters shown for 3 seconds runs at 20 CPS. Above roughly 20 CPS most adults start missing words; below it the read feels unhurried.

Treat 17 CPS as your working target and 20 CPS as a ceiling you cross only when the audio genuinely leaves no room. When a cue runs hot, you have two fixes: trim the text so there are fewer characters to read, or extend the out-point so there is more time to read them. You will use both. The subtitle word count tool reports CPS per cue and across the file, so you can see which lines are over budget before an editor ever complains.

How long should a line be?

The standard maximum is 42 characters per line, with at most two lines per cue. The limit exists because the eye reads a subtitle in a single glance without scanning left to right like a paragraph; a line much wider than 42 characters forces a scan and breaks that glance. Two lines of 42 is the practical ceiling of what a viewer absorbs before the cue changes.

When a thought needs more than 42 characters, split it across two lines at a grammatical boundary — after a clause, before a preposition, keeping articles with their noun. A line break in the wrong place ("the / red car") reads worse than a slightly uneven split ("the red / car"). If a cue would need three lines, it is too much text: cut it, or break it into two consecutive cues instead.

How short or long can a cue be?

Duration has a floor and a ceiling. The floor is about 0.8 to 1 second: below that, a cue flashes on and off before the eye can even register that it appeared, so a single word like "Yes." still needs roughly a second on screen. The ceiling is about 6 to 7 seconds: past that a static line starts to feel abandoned, and a viewer who has finished reading begins to wonder if it is stuck.

• Minimum ~1 second, even for a one-word cue, so it can be seen at all.
• Maximum ~6-7 seconds, after which a long line should be split into two cues that follow the speech.
• When a single spoken line runs long, break it into two timed cues rather than holding one cue for eight seconds.

These three limits interact. A cue that is at the reading-speed ceiling and also near the duration floor is a contradiction — too many characters, too little time — and the only fix is fewer words. Before you export, preview the whole track in the subtitle preview tool to feel the rhythm, and when you are ready to put the captions on the picture you can add subtitles to a video entirely in the browser. If the file needs to be a different format for its destination, the subtitle converter will reshape SRT, VTT or ASS without touching your timing.

Frequently asked questions

Do I have to type the timestamps by hand?

No. Paste your finished transcript into the text-to-subtitle tool and it generates the cue structure and timestamps for you to refine, so you are adjusting timing rather than typing "00:00:04,500" over and over.

What is a safe reading speed if I only remember one number?

Seventeen characters per second. Aim there and you will be comfortable for a general adult audience. Treat 20 CPS as the hard ceiling and only cross it when the audio truly leaves no room.

Should subtitles be word-for-word?

Rarely. Reading is slower than listening, so verbatim text usually outpaces what a viewer can read in the same window. Trim filler and false starts down to the meaning; that is what keeps the subtitle in step with the speech.

Is any of this uploaded to a server?

No. Every Pixora subtitle tool runs in your browser, so your transcript and timing stay on your own device and the file is never sent anywhere.