How to remove filler words from a video on iPhone

The reason filler words are painful to remove is not the cutting — it is the finding. On a waveform, “um” looks exactly like the word before it. You end up playing the take at half speed, ear tuned for a syllable you have already heard forty times, splitting the clip twice for each one.

A transcript changes the unit of work. When the audio has been turned into words, a filler word is just a token that a model can find, and you go from hunting to reviewing: the tool proposes every cut and you veto the handful you disagree with.

The fast way, on the phone

1. 01

   Bring the take into Primecut

   Open Primecut, create a project and pick the clip from your camera roll — or record it inside the app if you want the teleprompter. Nothing is uploaded to a social platform; the project lives on your phone.

2. 02

   Let it transcribe

   Primecut sends the audio for server-grade speech-to-text and comes back with a word-by-word transcript, usually in well under a minute for a short take. Every word is anchored to its exact position in the video.

3. 03

   Run the cleanup pass

   The AI pass marks filler words (um, uh, like, you know), dead air longer than a natural beat, and sentences you started over. Marked words show struck through in red so you can see exactly what is about to disappear.

4. 04

   Skim the marks and veto the ones you want to keep

   Tap any struck-through word to put it back. This is the step people skip and then regret: a deliberate pause before a punchline is not dead air, and 'like' inside a simile is not a filler word.

5. 05

   Export

   Choose 9:16, 16:9 or 1:1 and export. Free exports are 1080p with a small watermark; Pro exports up to 4K without one.

What to keep

Removing every filler word makes people sound like a press release. Three things are worth restoring on the review pass:

• The beat before a point lands. A pause you took on purpose reads as confidence; closing it up reads as rushing.
• Fillers that carry meaning. “Like” in “it was like a fire drill” is a simile, not a stumble.
• The breath at the start of a sentence. Cutting into the first consonant is the single most common cause of edits that “sound edited”.

Doing it without an AI editor

If you would rather stay in a general editor, the honest workflow is: export the audio, run it through a transcription service, note the timecodes, then split and delete in the editor. It works. It costs about ten to fifteen seconds per filler word, and it has to be redone from scratch every time you re-record. That trade is the whole reason transcript-based editors exist.

Questions