Procedure, start to finish
How to add subtitles to a client video online.
Six steps, one editing pass, one file to hand over.
Generating a subtitle track takes a minute. Making it good enough to put a client name on takes the pass that comes after, and almost nobody is taught the order to do it in. This page is the sequence: upload, generate, read against picture, fix the words, fix the timing, hand over the sidecar file.
Video hosting for filmmakers · 5 GB free · Paid plans from USD 9/month
Updated September 2026
The sequence
From a finished cut
to a file in the client’s hands.
This is the whole procedure in order. The parts people skip are step four and step five, and those are the two that decide whether the client thinks you delivered subtitles or thinks you forwarded a machine transcript. For why captions matter on client work at all, read the captions guide. This page assumes you already decided to do it.
Upload the locked cut
Upload the browser-playable export you were going to send anyway, in H.264, HEVC or AV1. Generate after picture lock, not before. A trim of two seconds in the middle moves every cue after it, and no amount of editing repairs that faster than generating again.
Check the minutes before you press the button
Generation is metered in minutes a month by plan, and the free Starter plan includes 20 minutes. Look at the runtime first. A 40-minute corporate film does not fit in a 20-minute allowance, and finding that out halfway through is the avoidable version of this problem.
Generate the track
A speech-to-text model transcribes the audio into timed cues. It is good at clean dialogue recorded on a lav, weaker on overlapping speech, room noise and heavy accents, and consistently wrong about anything it could not have heard before: your client name, their product, the town where it was shot.
Read it against picture, not on its own
Play the film with the caption panel open and watch the cues arrive. Reading the transcript in a text window hides the two failures that cost you the most: a cue landing on the wrong shot, and a line too long to read before it disappears.
Fix the words in a fixed order
Proper nouns first, then numbers and units, then punctuation that changes meaning, then line breaks and reading rate. The table below is that order, worst first. Do it in that order and you spend your time where a mistake actually embarrasses somebody.
Publish on the link, then hand over the sidecar
The edited track is the one the share link shows, so a reviewer watching the film sees the corrected version, and a reviewer who still spots a wrong word can submit a correction for you to apply. When it is right, download the file and send it with the delivery.
The editing pass
Five things to fix,
ordered by what costs you the most.
Every automatic transcript has the same failure profile, and the errors are not equally expensive. A misheard filler word costs nothing. A client company name spelled wrong in a file with your invoice attached costs a great deal. Work down this table and stop when the budget runs out, because the rows are already in cost order.
Worst first. Rows one and two are corrections of fact and are never optional on paid work. Rows three and four are craft. Row five is taste, and it is the first thing to cut when the deadline is real.
Timing
A cue can lead the line,
it should never lag it.
The second thing you fix is where cues sit. Automatic timing is usually close, and close is not the same as right. A cue that arrives a fraction late is worse than one that arrives a fraction early, and the reason is how people watch: the eye goes to text the moment it appears, then back to the face. A cue that is already there when the mouth moves reads as part of the film. A cue that appears after the words have been spoken makes the viewer feel they are being told what they just heard.
So when a cue sits slightly off, pull it earlier rather than pushing it later. A small lead is invisible. A small lag is not.
The other timing fault is a cue that outlives its shot. If a line stays on screen across a hard cut into a different scene, it attaches itself to the wrong image and the viewer reads it as a caption of what they are now looking at. Where a cut lands inside a sentence, end the cue on the cut and start the next one after it, even if that means splitting a line that would have read better whole.
Two habits save most of this work. Generate on the locked cut, so timings are never chasing an edit that moved. And do the timing check as a single continuous playback with the panel open, rather than scrubbing cue by cue, because drift is something you feel across a minute and never see in a list.
If you are going to search this footage later
The handover
Two files exist here,
and what to say to a request for anything else.
Captions download as .srt and .vtt, and only those two. That is not a tier limit or a setting somewhere; it is the complete list. Knowing that in advance is what lets you answer a format request in one line instead of going looking.
SRT, the default
Send this unless someone has told you otherwise. It is the format broadcasters, agencies, editors and nearly every platform expect as an input, and it is what an NLE imports when a client wants the text back inside their own timeline. If you send one file, send this one.
VTT, for the web
The format a web player expects alongside a video on a page. If the film is going onto the client site or into their own embed, they need this one, and sending it unasked saves a round trip with whoever builds the page.
Everything else
Not produced here. Hand over the .srt and say plainly that it is the source file to convert from. Almost every subtitle tool, editor and captioning service takes .srt as an input, so the conversion is a step on their side and not a gap in the delivery.
Send both files with the rest of the delivery rather than as a separate email later. A subtitle file that arrives a week after the video is a file that gets lost, and the client asks for it again on the day it is needed. Delivering final video files to clients covers how the rest of that handover is put together.
One caveat on the download. A vaulted master refuses caption download, because a subtitle file is a complete transcript of everything said in the film. If the material is under that level of protection, the transcript is protected with it, and the caption file is not something you can hand out from the vault copy.
Two honest limits
Burned-in text is an export setting,
and a runtime that fits the allowance.
If what the client actually needs is subtitles baked into the picture, this is not where that happens. Nothing is written into the image here, and the player serves the exact bytes you uploaded with no re-encode at either end, which is the same property that keeps your grade intact and the same property that means no text can be stamped on top of it. Burn the subtitles in your editor on export, style them there, and upload that export as its own film or as a new version of the existing one. Social cutdowns for silent autoplay are usually this case, and they are worth doing properly in the editor rather than approximately here.
The second limit is arithmetic. Generation is metered in minutes a month by plan, and the free Starter plan includes 20 minutes. A long film against a small monthly allowance is a problem you want to find before you start, so check the runtime against the allowance as step two of the sequence above and not as a surprise in the middle of it.
The short version
Questions
Frequently asked
What file formats can I download the subtitles as?
Two: .srt and .vtt. There is no third option here. If a client asks for anything else, hand them the .srt, which almost every converter, editor and platform accepts as an input, and let their side produce the format they need.
Can you burn the subtitles into the picture?
No. Nothing is ever written into the image here, because the player streams the exact bytes uploaded and no re-encode happens on the way in or out. Burned-in subtitles are an export setting in Premiere Pro, Final Cut Pro or DaVinci Resolve. Style them there, export, and upload that export as its own film or as a new version of the existing one.
How much can I caption per month?
Generation is metered in minutes a month and the allowance goes up with the plan. The free Starter plan includes 20 minutes a month. A feature-length film against a small allowance is the case to check before you start, not after.
Who can fix a wrong word?
The owner edits the track directly. A reviewer watching on the share link can submit a correction instead, and the creator applies it. That is the useful path for a client who knows how their own product name is spelled and you do not.
Why can I not download captions from this master?
A vaulted master refuses caption download. The Vault is the path for material that should not leave in file form, and a subtitle file is a complete transcript of the dialogue, so it is covered by the same refusal as the media. Vault access is included on every plan, including the free Starter plan.
Should I edit the transcript or watch the film?
Watch the film. A transcript read on its own hides the two errors that matter most: a cue that sits on the wrong shot, and a line that is technically correct but unreadable at the speed it appears. Both are only visible against picture.
Do subtitles change if I upload a new cut?
Treat them as tied to the cut you generated them from. If the edit moves, every cue after the change is off by the amount you moved it, so generate again on the locked version rather than nudging a whole track by hand.
Generate a track, then do the pass
Caption generation is metered in minutes a month by plan, and the free Starter plan includes 20 minutes. Check the runtime against the allowance before you start a long film.