A huge share of short video gets watched with the sound off — on the train, in bed next to someone sleeping, at work with the phone face-down on the desk. No captions means those viewers swipe past in two seconds, no matter how good your video is. Captions are not an accessibility nice-to-have; they are a retention tool, and they are free to add.

Our Auto Captions tool generates captions in the browser and burns them into your video with no watermark and no signup. If instead you need a separate subtitle file for YouTube or a course platform, the SRT Generator builds that. This guide covers when to use which — and how to style captions so they help instead of hurting.

Burned-in vs. SRT: which one do you need?

Burned-in captions become part of the video pixels. They show on every platform, in every player, with exactly the styling you chose — the bold animated word-by-word style you see on viral Shorts only exists as burned-in captions. The tradeoff: they are permanent. A typo means re-exporting, and viewers cannot turn them off.

SRT subtitles travel as a sidecar file. The viewer toggles them, the platform renders them in its own style, and you can fix a typo by editing text. The tradeoff: you depend on the player — some platforms ignore SRT on certain surfaces — and you give up all styling control.

The practical rule:

  • TikTok, Reels, Shorts → burned-in. These platforms are watched mostly on phones, often muted, and their players do not reliably offer subtitle toggles. Styled burned-in captions are the native language of short-form.
  • YouTube long-form, courses, client work → SRT (plus optional burned-in). Upload the SRT so viewers can toggle subtitles and YouTube can index the text; burn in captions as well if your style demands it, but keep them out of the lower third where YouTube’s own UI sits.
  • Ads → burned-in. You cannot rely on a viewer enabling subtitles on a paid impression. The caption is part of the creative.

The caption style guide for Shorts

Viral short-form captions share a visual grammar. Copy it:

Two to four words per line, centered. Never full sentences crawling across the screen. Short lines keep the viewer’s eyes in one place and create the punchy rhythm that holds attention. Break lines at natural phrase boundaries, not mid-phrase.

Big, bold, high-contrast type. The caption must be readable on a phone at arm’s length over a moving background. Heavy sans-serif (think TikTok’s classic style), white text with a thick black stroke or a solid background box. If a caption is hard to read for even a moment, the viewer is gone.

Highlight the keyword. Color one key word per caption group — the number, the product name, the punchline word. This is the signature move of high-retention captions: the eye jumps to the colored word, and the brain registers the point even while skimming. One highlight per screen; everything highlighted means nothing is.

Word-by-word or phrase-by-phrase timing. Karaoke-style word highlighting (each word lighting up as spoken) is the current short-form standard because it paces the viewer’s reading to your speech. Phrase-level timing is calmer and suits talking-head explainers. Never show a whole paragraph at once — it invites the viewer to read ahead and tune out your voice.

Safe zones. Keep captions in the vertical center band of a 9:16 video. The top is eaten by the platform’s UI and the creator name, the bottom by the caption text, like/comment buttons and progress bar. Center-screen captions survive every platform’s chrome.

Emoji, sparingly. One emoji per caption group at most, and only where it adds meaning (🔥 on the punchline, ⚠️ on the warning). Emoji-stuffed captions read as spammy and date fast.

Timing: the invisible half of captioning

Bad timing ruins good styling. The rules:

  • Captions lead the voice slightly. A caption appearing a fraction before the word is spoken feels instant; a caption lagging behind feels broken. Aim for the text to land within about 0.1–0.2 seconds of the speech.
  • Minimum on-screen time. No caption group should flash for less than about a second — viewers need time to register it. If your speech is very fast, merge short groups rather than flashing single words.
  • Cut on silence. Caption groups should break where you pause, not mid-sentence. This is one reason to remove dead air before captioning: clean pauses make clean caption breaks.
  • Punctuation matters. A missing question mark changes meaning; a missing comma changes rhythm. The auto-generation gets you close — the manual pass is where you add the punctuation that makes captions read like language instead of a transcript dump.

The correction pass: 5 minutes that separate pros from amateurs

Auto-captioning does the heavy lifting, but every transcript has the same weak spots. Run this checklist:

  1. Names and brands. Speech recognition mangles proper nouns first. Fix every product name, person name and brand — these are the words your video is about.
  2. Homophones. “Their/there/they’re,” “to/too/two” — the transcript picks the wrong one constantly.
  3. Jargon and slang. Niche terms, acronyms, and non-English words get creative reinterpretations. Search the transcript for anything that looks like a guess.
  4. Numbers. Prices, measurements, dates — verify every digit. A wrong price in a caption is a wrong price in your ad.
  5. Censorship false positives. Auto-captioning sometimes bleeps or stars out innocent words. Watch for asterisks that should not be there.

Five minutes of corrections on a one-minute video. That is the entire cost of professional-looking captions.

Captions and the algorithm: what is actually true

Platforms say captions help discovery because the text gives their systems more signal about the video’s content — and YouTube explicitly uses subtitle text this way for uploaded SRT files. Beyond that, the retention mechanism is straightforward and needs no study to understand: a viewer who can follow your video muted in the first three seconds stays; one who cannot, leaves. Since the opening seconds decide distribution, captions influence reach through retention even before any indexing effect.

What captions will not do: save a video nobody wants to watch. They are a multiplier on content that already works, removing a specific, common reason to swipe away. Apply the multiplier to everything — it costs nothing.

The workflow, start to finish

For a short-form video: edit first (including silence removal), generate captions with the auto captions tool, run the 5-minute correction pass, style them per the guide above — short lines, bold type, one keyword highlighted, center safe zone — and export. For YouTube or course content, generate the SRT file instead (or as well), upload it alongside the video, and let viewers toggle.

Then take the captioned video and multiply it: the same clip with captions performs across TikTok, Reels and Shorts without re-editing. If you are sitting on long footage — a product demo, a webinar, a podcast — cut it into clips with the free split tool first, then caption each one with the auto-caption tool. Caption once, post everywhere.