Three tips for
Reels people actually
watch to the end
How do auto captions work?
A speech recognition model listens to the audio track and produces a word-by-word transcript with a timestamp for every word (start and end, to the hundredth of a second). The software then splits that text into readable blocks and syncs them with the video.
A timestamped transcript does more than captions: it can also spot silences and filler words to cut them automatically, or highlight a specific word the moment it is spoken. For the visual side (styles, karaoke, keyword highlights), see our animated captions guide. This page covers generation, accuracy and file formats.
Step by step
- Get clean audio
Record with a mic close to your mouth, no background music, in a furnished room.
- Set the language
Choose the spoken language instead of letting the tool guess, especially for short clips.
- Run the transcription
The software transcribes speech word by word and timestamps each word.
- Proofread the risky parts
Names, numbers, homophones and punctuation, while playing the video at 1.5×.
- Pick the format
Burned-in for Reels, TikTok and Shorts; an SRT file for YouTube, websites or course platforms.
- Finish for accessibility
Add meaningful sounds in brackets and identify speakers where needed.
| Criterion | SRT / VTT file | Burned-in |
|---|---|---|
| Visible by default | No (viewer turns on) | Yes |
| Can be turned off | Yes | No |
| Multiple languages | Yes, one file per language | One video file per language |
| Styling and animation | No (plain text) | Yes |
| Reels, TikTok, Shorts | Poor fit | Standard |
| Long YouTube videos, websites, courses | Recommended | Possible |
| Editing after publishing | Easy (upload a new file) | Re-export needed |
| Accessibility | Good if proofread | Good if proofread and readable |
How accurate are auto captions?
With a clear voice recorded close to the mic in a widely spoken language, mistakes are rare: a few words per minute at most. Accuracy drops in predictable situations:
- Distant or echoey audio: an empty room, a phone placed 10 feet away.
- Music or noise under the voice during recording.
- Overlapping speakers.
- Names and brands: product names, surnames and street names are often misspelled.
- Numbers, units and acronyms: "$2.50", "ROI", "sq ft".
- Code-switching: mixing two languages in one sentence.
- Strong accents and very fast speech.
The biggest improvement is free: use a lavalier mic, or hold the phone 12 to 20 inches from your mouth in a furnished room. That will do more for accuracy than switching tools.
How to proofread auto captions in 2 minutes
For a one-minute video, a targeted proofread rarely takes more than two minutes. Check in this order:
- Proper names: people, companies, places, products.
- Numbers: prices, dates, percentages, phone numbers. A mistake here can be costly.
- Homophones: "their / there", "your / you're", "fifteen / fifty".
- Punctuation: a misplaced period changes how blocks are split.
- Words split across two blocks, and blocks too short to read.
Tip: play the video at 1.5× and read only the text. Errors jump out much faster than when reading the transcript alone.
SRT file or burned-in captions: which should you use?
There are two ways to deliver captions, and they serve different purposes.
Closed captions in a separate file (SRT, VTT)
An SRT file is plain text: a number, a timestamp such as 00:00:01,200 --> 00:00:03,400 and one or two lines. Viewers turn captions on or off, and you can offer several languages. YouTube accepts basic SRT files in plain UTF-8 without styling (YouTube Help: supported caption files). This is the right choice for long videos, websites, course platforms and any player that supports captions.
Open captions (burned-in)
The text is part of the image: it shows everywhere, with no action from the viewer, in your font, colors and animation. It is the standard on Reels, TikTok and Shorts, where most people never touch the captions button. The trade-off: they can't be turned off and there is one language per video file.
Many creators do both: a burned-in version for social media, an SRT file for YouTube and their website.
Readability rules for classic captions
Caption files follow professional standards. Netflix's English style guide, for example, sets 42 characters per line, a maximum of 2 lines and up to 20 characters per second for adult programs (Netflix Timed Text Style Guide); many other languages use 42 characters per line with a slower reading speed. For burned-in vertical captions, go much shorter: 2 to 5 words per screen.
Break lines at natural points (not between "the" and its noun) and keep each caption on screen for roughly a second or more.
Auto captions and accessibility
For deaf and hard-of-hearing viewers, captions are essential. The Web Content Accessibility Guidelines (WCAG, success criterion 1.2.2, level A) call for captions on prerecorded video with audio. Accessible captions go a step beyond a transcript:
- describe meaningful sounds:
[laughter],[music],[doorbell]; - identify who is speaking when several people talk;
- proofread: uncorrected auto captions are not enough for corporate, public-sector or commercial video.
Public bodies and some businesses have legal accessibility obligations depending on the country. If that applies to you, check the exact requirements for your videos.
Languages and mixed-language videos
Always set the spoken language before transcribing: models that auto-detect the language get it wrong more often on short or bilingual clips. For a video that switches languages, transcribe each part in its own language or proofread the mixed passages more carefully.
Orbicut transcribes 30+ languages automatically, including English, French, Dutch, German, Spanish and Italian, with a timestamp for every word. Tap a word to fix it, and the captions are burned in when you export. Transcription is included on every plan, the free one too, and uses no credits (see pricing). To reach viewers in other languages, read our guide on how to translate a video.
Frequently asked questions
What is the difference between captions and a transcript?
A transcript is the full text of what is said. Captions are that text split into short blocks synced with the video. Both come from the same speech recognition.
Instagram and TikTok already offer auto captions. Why add my own?
Native captions depend on the app and on each viewer's settings, with little control over style and position. Proofread burned-in captions look the same everywhere and can match your brand.
Can I generate captions for a video without speech?
No, automatic transcription needs speech. For music-only or silent videos, write on-screen text manually as titles or cards.
How long does it take to auto caption a video?
Transcribing a few minutes of video usually takes under a minute. Plan for about two minutes of proofreading per minute of video for a reliable result.
Are closed captions or open captions better for SEO?
A caption file gives platforms like YouTube readable text alongside the video, which burned-in captions do not. For long-form YouTube content, upload an SRT file even if you also burn captions into your social clips.