Automatic captions are literal. They write down what you said, which is the whole point, right up until the moment the thing you said is a word you did not want printed across the middle of your video in bold.

Editing it out by hand works once. It stops working when you post daily, or when a word you use constantly is one you never want on screen.

Why this needs a rule, not an edit

A caption track is not one block of text. It is a sequence of cues, each holding individual words with their own timings, which is what makes word-by-word highlighting possible in the first place.

A video frame with captions burned in, one word highlighted as it is spoken.

That structure is why hand-editing is tedious. The word you want to mask appears in several cues, each time as its own timed token, and you have to find every one. Regenerate the captions after any change to the cut and they all come back.

A rule survives regeneration. You name the word once, and every caption pass applies the rule on the way out.

Captioning like this leans on Apple's Speech framework, which supports recognition performed locally - the reason the audio never needs to be sent anywhere to be read.

Treat captions as access first and engagement second. The standard most teams are held to is the W3C's captions for prerecorded media.

What the masking actually does

In Zella on iPhone, transcript rules live in Studio under Captions. Add a word to the censor list and it renders with its first letter intact and the rest replaced by asterisks, so "damn" becomes "d***".

Three details make the difference between a rule that works and one that misses half the time:

  • Case does not matter. The list matches regardless of how the word was transcribed.
  • Punctuation does not break it. A word transcribed with a trailing comma or full stop still matches, because the match is made against the bare word.
  • It applies at generation and afterwards. Newly generated captions come out already masked, and you can apply the rules to a caption track that already exists rather than regenerating it.

Keeping the first letter is a deliberate choice. Fully replacing a word leaves the viewer wondering what was said. Keeping the initial preserves the reading rhythm and makes the mask legible as a mask.

The part people get wrong

Censoring the caption does not censor the audio.

This sounds obvious written down, and it still catches people out, because the word disappears from the part of the video they are looking at while they work. The spoken word is untouched in the audio track. Anyone with the sound on hears it exactly as recorded.

If you need it gone from the audio as well, that is a separate edit: cut the word out on the timeline, or drop the clip volume over it. Do that first, then let the caption rules handle the text, because cutting audio changes the timings that captions are built from.

For a platform that penalises certain words in speech, only the audio edit counts. For a platform that reads your captions or your uploaded subtitle file, the text rule is the one that matters. Most creators need both.

The two rules that ship alongside it

The same panel carries two others, because they solve the same category of problem:

Capitalize sentences puts a capital on the first word of each cue. Speech recognition often returns lowercase runs, and a caption that starts mid-sentence with no capital reads as sloppy at a glance.

All caps does the opposite and shouts everything, which is the loud reel style. It is a style choice rather than a hygiene one, and it deliberately overrides sentence capitalization when both are on, because there is no such thing as a capital in a line that is already entirely capitals.

A workflow that holds up

The order that avoids rework:

  1. Lock your cut first. Every caption timing derives from it.
  2. Handle any word you need gone from the audio on the timeline.
  3. Generate captions.
  4. Add your censor words and capitalization rules, and let them apply to the track.
  5. Style last: pick the caption look once the text has stopped changing.

If you are working in more than one language, note that the censor list is a list of literal words. It masks what you put in it, in the language you put it in, so a bilingual video needs both spellings. That is also true of English subtitles generated from another language, where the words on screen are English regardless of what was spoken.

Frequently asked questions

Does censoring the caption also bleep the audio? No. The rule changes the caption text only. The spoken word is still in the audio and needs a separate edit on the timeline.

What does a censored word look like? The first letter, then asterisks for the rest, so a four-letter word becomes one letter and three stars.

Do I have to add every capitalization of a word? No. Matching ignores case, and it ignores punctuation attached to the word.

Will the rule survive regenerating the captions? Yes. The rules are stored on the project and applied at every caption generation, and you can also apply them to a track that already exists.

Can I use it to swap a word rather than mask it? Not with the censor list. It masks a word rather than substituting a different one. Edit that cue directly if you want different wording.

Does ALL CAPS override the capitalize rule? Yes, deliberately. A fully uppercase line has no separate first-letter capital to set.