Subtitles: SRT, VTT, and what actually helps a video

Subtitles are one of the few things in video production where doing the minimum and doing it well cost almost the same, and the gap in outcome is large. Most channels do neither: they leave automatic captions on and never look at them, or they burn in a caption track positioned where the platform’s own interface covers it.

Here is what the choices actually are.

The two formats, and why it barely matters

SRT — SubRip. Numbered cues, timestamps with a comma before the milliseconds:

1
00:00:01,000 --> 00:00:04,120
This is the first line of narration.

VTT — WebVTT. A WEBVTT header at the top, no cue numbering required, a full stop before the milliseconds:

WEBVTT

00:00:01.000 --> 00:00:04.120
This is the first line of narration.

That is essentially the whole difference for ordinary use. VTT supports styling and positioning that SRT does not, and it is the format HTML video elements expect. SRT is what most editors export and what most platforms accept without complaint.

Practically: use SRT unless something specifically asks for VTT. If a tool rejects your file, converting between them is a text transformation, not a re-render — the subtitle converter does it in the browser, which also means a caption file containing unreleased content never leaves your machine.

The same conversion the other direction — stripping the timings to plain text — is more useful than it sounds. It gives you the transcript, which is the fastest route to a video description, a written version of the video, or an accurate word count for a script you have already recorded.

Uploaded file or burned in?

This is the decision that actually matters, and the right answer differs by format.

Long-form: upload the file. A caption file is:

• switchable, so viewers who do not want captions can turn them off
• translatable, either by the platform or by you supplying additional files
• searchable, giving the platform an accurate transcript rather than a machine guess
• correctable after publishing, without re-exporting anything

Burned-in captions are none of those things. They are pixels. A typo is permanent, and a viewer who finds them distracting has to leave.

Vertical short-form: burn them in. The reasoning inverts. Most short-form viewing is muted, the player’s own caption rendering is inconsistent, and the visual rhythm of the captions is part of the edit. Here burned-in text is the norm and it is the right call.

But burn them in in the right place, which is where most people go wrong: not the lower third. In vertical the bottom of the frame belongs to the channel name, the caption, the progress bar and often a music strip. Subtitles placed there compete with the platform’s own text. Slightly above the vertical centre, with lines ending well clear of the right-hand action rail, is the working position — you can check yours against the real overlay with the Shorts safe zone tool.

Getting a transcript in the first place

If you have the script, you already have the text — but not the timings, and the timings are the work.

If you do not have the script, or the narration drifted from it during recording, transcribe the finished audio. The video transcriber takes a file you already have and returns text with timings, exportable as SRT, VTT or plain text. Either way, the machine output is a draft, not a deliverable.

Automatic transcription is now good enough that people stop reading the output, which is exactly when it hurts. It reliably fails on:

• Proper nouns. Names, brands, product names, place names — the words most likely to be the point of the sentence.
• Numbers and units. “$1.5M” versus “one point five million” versus “1.5 m”.
• Technical vocabulary. Any term specific to your niche is the term most likely to be wrong.
• Homophones under accent. Whole sentences that parse as English but say something else.

The fix is a single read-through with the audio playing. It takes a fraction of the video’s runtime and it catches the errors that are most visible, because mis-transcribed proper nouns are precisely the words a viewer is watching for.

Formatting that makes captions readable

Captions that exist are not the same as captions that help. A few constraints do most of the work:

• 32 to 42 characters per line. Longer lines get truncated or wrap awkwardly on phones, where most viewing happens.
• Two lines maximum on screen at once. Three-line blocks read as a wall and pull the eye away from the picture.
• One to six seconds per cue. Under a second and it flickers; over six and the viewer has finished reading and is waiting.
• Break at clauses, not at the line limit. “The problem with this approach / is that it never scales” reads correctly. “The problem with this approach is / that it never scales” makes the viewer re-parse.
• Do not caption filler. “Um”, “you know”, false starts. Transcribe what was meant, not every noise. This is standard practice, not dishonesty.
• Keep the reading speed sane. If narration is running well above normal pace, captions become a speed-reading exercise. Condensing slightly is better than perfect fidelity nobody can follow.

If you are unsure whether a line is too long, character counts are the whole test — the word and character counter is faster than eyeballing it.

Chapters are the other half of this

Captions make a video followable while it plays. Chapters make it navigable, which is the thing that keeps someone from leaving when the section they want is four minutes in.

They cost about two minutes to add: timestamps in the description, first one at 00:00, each one labelled by what is in it rather than “Part 2”. The timestamp generator formats them correctly, which matters because platforms are strict about the format and silently ignore chapters that are slightly wrong.

Chapters help most on exactly the videos where retention is hardest — longer explainers where a viewer arrived for one specific section. Without them, that viewer either scrubs blindly or leaves.

A sequence that takes about fifteen minutes

1. Transcribe the finished audio, or start from the script if the narration matched it.
2. Read the transcript once with the audio playing, fixing names, numbers and jargon.
3. Reflow the lines to 32–42 characters, two lines maximum, breaking at clauses.
4. Export SRT and upload it, for long-form. For vertical, burn in — above the midpoint, clear of the right rail.
5. Pull chapters from the section headings you already have and paste the timestamps into the description.
6. Keep the plain-text transcript. It is the fastest first draft of a written version of the video, and a genuinely useful description.

None of this is glamorous, and none of it will save a video that nobody wanted to watch. But it is close to the highest ratio of outcome to effort available in post-production, and it is one of the few parts of the job where “did it properly” and “did it at all” take nearly the same amount of time.