1. What this lesson is for

Auto-captions get you 80-95% of the way in seconds. The last few percent - misheard words, broken names, awkward line breaks, captions that flash too fast - is what separates a professional video from one that looks careless. In this lesson you turn auto-generated captions into clean, accurate subtitles ready for publication.

Keep the module principle in mind: readability beats decoration. A plain, correct, well-timed subtitle beats an animated one with a typo in the key word.

2. Caption generation settings

CapCut offers automatic captions from the audio on the timeline. Where exactly the feature lives (a Text or Captions panel, or a right-click option on the clip) and which options it shows differ between desktop, mobile and web, and between versions. Look for these settings before you generate:

• Spoken language - set it explicitly instead of relying on auto-detect, especially for mixed-language speech.
• Source - generate from the voice track only if possible. Music and sound effects under the voice cause misheard words.
• Filler word removal - some versions can drop "um" and "uh" automatically. Useful, but check that it did not remove meaningful words.
• Style - pick a plain style first. Apply your branded subtitle preset (lesson 2.3) after the text is correct, so styling never hides errors.

3. The correction workflow

Correct in separate passes. Mixing them makes you miss errors.

4. Common punctuation and spelling fixes

• Proper names and brand terms - the most frequent failure. Check every name, product and technical term against a term list.
• Homophones - "their / there", "your / you're", "four / for".
• Numbers - decide once: digits ("3 tips") for short-form video is usually clearer than words.
• Punctuation style - pick one rule for final periods (many short-form creators drop them on single-line captions) and apply it everywhere. Inconsistent punctuation is a listed pitfall of this module.
• Casing - sentence case reads fastest. ALL CAPS only for a single emphasized word, if at all.

Keep a term list per channel or series so you fix each word once, not every video:

5. Line breaks and chunk length

A caption chunk is what is on screen at one time. Short-form vertical video works best with short chunks: one line, or two short lines, of roughly 32-42 characters per line at most. Many creators go even shorter (3-6 words) for a punchy rhythm.

• Break at natural phrase boundaries: after a comma, before "and / but / so", between subject and verb phrase only if unavoidable.
• Never split an article from its noun ("the / camera") or a first name from a last name.
• Keep the chunk to one idea. If a chunk contains two ideas, split it into two chunks.
• Avoid a very long first line over a one-word second line - balance the lines.

BAD                                   GOOD
"So the first thing you need to"      "So the first thing"
"do is check the"                     "you need to do"
"audio levels before export."         "is check the audio levels"
                                      "before export"

6. Keyword highlighting and timing

• Reading speed - aim for about 15-17 characters per second as a comfortable upper limit for most viewers; above roughly 20 cps captions become hard to follow. Give each chunk at least about 0.8-1 s on screen.
• Sync - the chunk should appear when the first word is spoken and leave when the last word ends or shortly after. Watch for a constant offset (all captions early or late by the same amount) and fix it by shifting the whole caption track, not each chunk.
• Gaps - during pauses longer than about 1 s, let the screen be empty rather than holding the old caption.
• Keywords - highlight at most one word per chunk, and not in every chunk. If everything is highlighted, nothing is. Use one highlight color from your brand kit.

7. Caption QA log and checklist

Log every correction. Over a few videos the log shows you which errors repeat.

ERROR LOG - series: capcut_tips  (after 5 episodes)
RECURRING MISHEARD WORDS : "cap cut" (5/5), "be roll" (3/5), "keyframe" -> "key frame" (2/5)
PROPER NAME FAILURES     : host name "Artur" -> "Arthur" (4/5)
TIMING OFFSETS           : ep03 whole track +0.3 s (music under voice), ep05 none
ACTION                   : add terms to term list; generate from voice-only track

CAPTION CHECKLIST
[ ] Reading speed under ~17 cps, every chunk on screen at least ~1 s
[ ] Punctuation and casing normalized to the series rule
[ ] All names and terms match the term list
[ ] Line breaks at phrase boundaries, one idea per chunk
[ ] Highlighted keywords used sparingly (max one per chunk)
[ ] Captions sit inside the safe area, above the bottom UI

8. Exercise

1. Take a finished rough cut of 45-90 seconds and generate auto-captions from the voice track with the language set explicitly.
2. Do the four passes in order: words and punctuation, line breaks, timing, then styling.
3. Fill in a caption QA table with at least five corrections and start a term list with at least five entries.
4. Check reading speed on the three densest chunks and fix any that exceed ~17 cps.
5. Watch the export on a phone with sound off. Could someone follow the whole message from the captions alone? Note what still needs fixing.

Further reading

• CapCut Help Center
• BBC Subtitle Guidelines
• Netflix - English timed text style guide
• W3C WAI - Captions/Subtitles