Captions for Reels and Shorts that survive the upload.

The first second is the whole problem

Reels and Shorts autoplay without sound. That means the opening moment of your clip is silent by default, and whether a viewer stops scrolling depends entirely on what they can see. Captions are not an accessibility feature in that window — they are the only channel carrying what you are saying.

Which reframes what "good captions" means for this format. Accuracy still matters, but legibility at a glance matters more, because a perfectly accurate caption nobody can read in the time it is on screen did nothing. Most caption mistakes on vertical video are styling mistakes, not transcription ones.

The practical consequence: style for the worst frame of your clip, not the average one. Scrub to the brightest, busiest second and judge your captions there. If they hold up over that, they hold up everywhere.

Why a subtitle file does not work here

You can attach a .srt file to a video in plenty of tools. On short-form social it will frequently not reach your viewer.

Platforms re-encode what you upload into their own formats. A sidecar file is not part of the video stream, so unless the platform explicitly supports uploading and retaining one, it is simply not carried. Then there is everything downstream: someone saves your clip and reposts it, sends it in a DM, screen-records it, or it appears in an embedded player. None of those carry your subtitle file.

Burned-in captions — the words rendered into the pixels — survive all of it. They cannot be switched off and cannot be stripped. For this format that is not a preference; it is the only approach that reliably puts your words on screen.

The real cost is that burning in is irreversible, which is why it matters that captions stay editable data until the moment you export, and why you can render the same source again with different text or styling as a separate file.

Placement on a 9:16 frame

Vertical video is not just a narrow 16:9. The platform puts its own interface over parts of your frame, and that interface is not in your preview.

The bottom of a Reel carries the caption text, the audio attribution and the action buttons; the right edge carries likes, comments and share. Captions sitting in the lower third — the natural position inherited from widescreen video — are exactly where they are most likely to be covered or crowded.

The usual answer is to lift the caption block toward the vertical middle, which also happens to be closer to where a viewer's eye already is. CaptionFX exposes caption placement so you can move the block and preview it against your actual footage, rather than discovering the collision after posting.

Worth checking against your own on-screen text too. A caption that clears Instagram's interface but sits on top of your own title card is the same problem.

A styling recipe that works

If you want a starting point rather than a theory: a heavy weight, a strong outline, size generous enough to read at arm's length on a phone, short-phrase or word-by-word pacing, and one highlighted word per line.

The outline is the part people skip and it is the one doing the most work. Your background is moving and you do not control it — a caption crosses a bright sky, then a dark jacket, then a white wall within two seconds. An outline guarantees a contrast edge whatever is behind the letter. A thin light font with no outline looks superb in a design mockup and is the single most common caption mistake there is.

Emphasis earns its place when it is rare. One or two highlighted words per line gives a scrolling viewer somewhere to look and carries the gist to someone reading only those. Highlight half the line and you have highlighted nothing, and made the text harder to read.

Where the errors will be

Clean English audio comes back around 95% of words correct, which on a 45-second clip is a handful of wrong words. They are not randomly placed — they cluster on names, brands, numbers and jargon, because those are the words a speech model has the least evidence for and the ones your audience notices first.

So the fast review is to check those categories rather than read every line evenly. Do it with the video playing, which also catches the error a transcript cannot show you: a caption that is correct but on screen too briefly to read.

If you are recording in Hindi or Hinglish, expect more. Clean Hindi runs around 85-90%, and code-mixed speech is below both — the model has to decide language and script word by word. It is a review pass, not a problem, but budget for it.