Each caption is drawn onto the frames it belongs to and the video is encoded once, so the words become part of the picture. Load an SRT or VTT file, or write the lines yourself with a timestamp above each one. The preview shows a real frame with the caption on it, so the size and position can be judged before anything is encoded.
Burnt-in captions show everywhere: in a muted feed, in a player with no subtitle button, and after the video has been downloaded and reposted. A separate subtitle track is better where it is supported, because it can be turned off and translated — but most social platforms do not show one at all, which is why the words are put into the picture here. This does not listen to the video; the words come from you, because automatic transcription needs a speech model far too large to load in a browser.
Captions people can actually read
-
Every line gets an outline, whatever else you choose:
white text alone disappears the moment the footage behind it turns pale, and that is the usual reason burnt-in captions fail. The outline costs nothing and works over anything.
-
Feeds play silently, so captions carry the message:
most social video autoplays muted, which means a clip that depends on its audio to make sense already fails for most of the people who see it. Captions are not an accessibility extra there; they are the content.
-
Size for a phone, not for your monitor:
text that looks generous on a laptop is small on a handset held at arm’s length. If the video is for a feed, the large setting is usually right even though it looks oversized while you are editing.
-
Keep lines short:
a caption is read in a couple of seconds while something else is happening on screen. Lines are wrapped to fit the frame, and a caption that wraps to four lines is one that should have been split into two captions.
Why burned in rather than a track you can toggle
A separate subtitle track is the better technology where it is supported: it can be switched off, translated, and styled by the player. The trouble is that most social platforms never show one at all, so a caption sitting in a track nobody renders helps nobody. The [msm_tool slug="video-subtitles"] draws each line directly onto the frames it belongs to instead, and encodes the video once with the words already part of the picture — which is why they show up in a muted feed, in a player with no subtitle button, and after the clip has been downloaded and reposted somewhere else entirely.
It will not listen to the video for you
Turning speech into text needs a model far larger than a browser can load, so the words have to come from an SRT or VTT file, or be typed by hand in the same shape: a timestamp line, an arrow, an end time, then the words. Both a comma and a full stop are accepted before the milliseconds, since different tools export the format slightly differently.
The outline is not a style choice
Every line is drawn with an outline regardless of what else is set, because plain white text disappears the moment the footage behind it turns pale — the ordinary reason a burnt-in caption becomes unreadable. The outline costs nothing to render and holds up over any background, which is the point of it not being optional.
Size for the phone it will actually be watched on
Feed video autoplays muted on almost every platform, which makes a caption the content rather than an accessibility extra — if a clip depends on its audio to make sense, most people who see it never hear it. Text that looks generously sized on an editing monitor reads small on a handset held at arm's length, so the large setting is usually the right call even though it looks oversized while you are working on it.
What a caption running past the end is telling you
A warning that the last caption ends after the video does almost always means the timing file was written for a different cut of the footage — a version before it was trimmed, or a different export entirely. The line stating when the final caption lands is worth checking against the video's actual length before encoding.
Once it is burned in, it stays
There is no undo for this the way there is for a subtitle track: the captions become part of the picture and cannot be hidden or translated afterwards. Keep the original video file and the subtitle file separately, since running this a second time on the burnt-in output would draw a fresh layer of text over the first.