How to stop the picture showing someone mouthing words you cannot hear

Last updated 2026-09-15 · 8 min read

What is the short version?

Most of a talking-head film is cut to the words, so mouths and audio agree by construction. The failure happens at the joins — a line ends a beat before the shot does, and for half a second the picture shows someone speaking into silence.

1. Ask Vizard Agent to audit every talking shot against its audio.
2. Ask it to find windows where the mouth is genuinely still.
3. Ask for blinks and held expressions checked as well.

What do you need before you start?

The cut and the rushes behind it. Vizard Agent's best fix is a different window of the same person — listening, nodding, reacting — and those only exist if it can look through everything that was filmed rather than what was used.

• The cut. The version with the problem.
• The rushes. All of them, including unused takes.
• Which person. If there is more than one.
• What can be covered. Whether b-roll is acceptable.
• How strict. A campaign film allows no slips at all.

What do you type into Vizard Agent?

Describe the symptom precisely. Telling Vizard Agent that the lip sync is off sends it hunting for a timing offset, when what you actually have is the opposite of that — sync that is correct everywhere, and picture that simply outlasts the audio underneath it.

Prompt

Variants worth knowing:

• "Mouth moves without audio under it." The actual fault.
• "Windows where her mouth is closed and still." The fix you want.
• "Frame by frame." Half a second is enough to notice.

What does Vizard Agent actually do?

Here is the Vizard Agent sequence on a real campaign film where silent mouth movement had to be eliminated completely. The audit is what finds the moments, and most of the work afterwards goes into hunting through the rushes for windows that can replace them.

1. Runs a frame-accurate lip-sync audit alongside the transcript.
2. Reads the audit results to list every offending moment.
3. Examines the mouth through a specific insert shot.
4. Checks whether the lips move during that insert at all.
5. Grabs a frame from every original clip and builds contact sheets.
6. Surveys every usable clip of that person.
7. Scans for a mouth-closed, stable window.
8. Finds the quietest mouth window among the candidates.
9. Checks for held blinks as well as mouth state.
10. Confirms the chosen window by looking at the lips closely.

Step nine is the one that separates a good fix from a nearly-good one. A window where the mouth is still but the subject holds a blink for a second looks just as wrong, and it will be the thing a client cannot name but does not like.

Step seven is not the same as finding a nice frame. A still frame proves nothing, because the mouth has to stay still for the whole duration you need, and the window is what Vizard Agent is measuring.

Step one covers the film rather than the moment you reported. Silent mouth movement clusters at cut points, so where there is one there are usually two more.

What does the result look like?

Picture that never outruns its audio. Where the shot had to change, the new window is the same person in the same setting, listening rather than speaking, and Vizard Agent tells you every moment the audit found and what it used at each one.

When does this not work well?

Sometimes nobody is ever still. A subject who talks continuously through every take leaves Vizard Agent no closed-mouth window to find, and at that point the honest options are a cutaway, a tighter shot that excludes the mouth altogether, or simply ending the shot earlier than you wanted to.

• A subject who never stops talking. No window exists.
• Single-take footage. Nothing else to cut to.
• Very long windows needed. Stillness rarely lasts.
• Close framing on the mouth. Nothing can be excluded.
• Everyone in frame is talking. A group makes it harder.

How do you fix a result that came back wrong?

Give the timecode of what you saw. Vizard Agent has the audit, so it can compare what you noticed against what it flagged — and a moment it missed tells it something about how the audit was set.

• "It still happens at eleven seconds." That moment is re-audited.
• "She blinks oddly in the new shot." Another window is found.
• "The replacement does not match." Framing and light are compared.
• "I would rather cover it." A cutaway is used at that point instead.

How does Vizard Agent compare to doing it yourself?

By hand you fix the moment somebody flagged, and the same fault is three more places in the film that nobody has watched closely enough to notice yet. Finding them means watching a thirty-second film about twenty times.

Common questions

Is this a lip-sync problem? No, and saying so misdirects Vizard Agent. The sync is correct; the picture outlasts the audio.

Can it find every instance? Vizard Agent audits the whole film rather than only the moment you report.

Is b-roll not simpler? Sometimes. A listening shot of the same person usually reads better.

What is a mouth-still window? A stretch long enough for the gap, not a single frame. Vizard Agent measures the stretch.

Why do blinks matter? Because a held blink looks as wrong as a moving mouth, and Vizard Agent checks both.

What if there are no still windows? Vizard Agent says so and offers a cutaway or an earlier cut instead.

Does the audio change? No. Vizard Agent changes only the picture over those moments.

Will the replacement match? Vizard Agent compares framing and light before using it.

Can it check a longer film? Yes, though the audit takes longer. The method is the same.

What about a two-person shot? Harder. Tell Vizard Agent who is speaking and it audits accordingly.

Can it just end the shot earlier? Yes, if the timing allows. Vizard Agent will tell you what that costs.

Does this apply to dubbed video? Differently — there the mouth disagrees throughout, which is a separate job.

How long a gap is noticeable? Half a second is plenty. That is why the audit is frame accurate.

Can I see the audit? Yes. Ask Vizard Agent for the list of moments and what it did at each.