Subtitling looks like transcription with timestamps. It is not, and the people who treat it that way produce files that get sent back.
The words were the hard part for about fifteen years. They are not any more. What remains is a set of judgements software has no basis for making: when a line should appear relative to the speech, how long a viewer needs to read it, where to break a sentence so it reads as a phrase rather than a width, and what to cut when a line is too long to be read at the pace it is on screen.
What it pays
Rates are quoted per video minute, not per hour of your time, and the two are far apart.
| Work | Per video minute | Real hourly, roughly |
|---|---|---|
| Correcting auto-caption timing | $1 – $5 | $12 – $30 |
| Subtitling from nothing | $3 – $8 | $15 – $35 |
| Translating existing subtitles | $4 – $12 | $18 – $45 |
| Captions for deaf and hard-of-hearing viewers | $4 – $10 | $18 – $38 |
A usable pass takes three to five times the length of the video. A ten-minute video is thirty to fifty minutes of work, and anyone quoting per video minute is quoting a number you have to divide before comparing it to anything.
Why timing is the whole job
A line has to arrive with the speech
Not after it. A subtitle a beat late reads as a mistake even when the words are right, and auto-captions do this constantly.
It has to stay long enough to read
Seventeen characters a second is the working figure. Under that and viewers miss it; well over and the subtitle is still there when the speaker has moved on.
It has to break where sense breaks
"I went to the shop and" / "bought milk" reads worse than "I went to the shop" / "and bought milk", though both fit the width. Software breaks on width.
Two lines, never three
A third line covers picture and pushes reading time past what the shot allows. When two will not fit, the answer is fewer words rather than another line.
What automatic captions actually leave you
Run any video through auto-captioning and the transcript will be close to right. Then watch it. The errors that remain cluster in exactly the places a human was needed: proper nouns, technical vocabulary, crosstalk, anything said quietly, and every single timing decision.
That is why most paid subtitling work now arrives as a correction task with a draft attached rather than a blank file. It pays less per minute than subtitling from scratch and takes considerably less time, so the hourly rate is often better.
Captions and subtitles are different deliverables
Subtitles assume the viewer can hear and renders speech only. Captions for deaf and hard-of-hearing viewers include sound events — a door, music starting, laughter — and identify speakers when it is not obvious who is talking.
Being asked for one and delivering the other is the most common reason a first job comes back. The brief should say which; if it does not, ask before starting rather than guessing.
Where the work is
Subtitling sits in the same market as transcription and often at the same clients, but the pool of people who can do it well is smaller because the timing work discourages anyone looking for something purely mechanical.
On Brief Buck it appears as a bounded task with the video, the format and the reading-rate target stated up front, and the file as the deliverable. The subtitling task page sets out what a poster is asked to specify, which is also the list worth checking before you accept work anywhere else: the format, the frame rate, whether sound events are wanted, and what the reading-rate target is.
Who this suits
It suits people who are patient with detail and content to work slowly. It does not suit anyone hoping for speed: the rate per video minute looks generous until you have timed a ten-minute file for forty minutes.
If you would rather type than time, transcription is the adjacent work and pays on a different basis. If you would rather speak than type, voice-over work and audiobook narration are the other direction entirely.