How to cover a re-recorded line with b-roll so the lips do not give it away

Last updated 2026-09-10 · 8 min read

What is the short version?

Replacing a phrase in a cloned version of your own voice is the easy half. The hard half is that the face on screen is still forming the original words, and a viewer notices that even when they cannot say why.

1. Go to Vizard Agent with the corrected script.
2. Ask it to cover the replaced lines with b-roll.
3. Ask it to verify the cover reaches both ends.

What do you need before you start?

The recording and a clear list of what changed. Vizard Agent needs the replaced passages as timings rather than as descriptions, because "the bit where I stumbled" is not something a cutaway can be placed against with any precision.

• The original recording. Picture and sound.
• Which lines were replaced. With their timings.
• B-roll, or permission to find some. It has to cover something.
• How the cutaways should feel. Illustrative, or neutral.
• Whether any face-to-camera moments must survive. Say so.

What do you type into Vizard Agent?

Ask Vizard Agent for cover rather than for correction. The instruction that actually matters here is that the picture must leave your face for the whole of those seconds, and that Vizard Agent should go back and confirm it did rather than assuming the placement worked.

Prompt

Variants worth knowing:

• "Exactly those passages." Not roughly around them.
• "Check the render." Placement is not proof.
• "Both ends." The tail is where it usually fails.

What does Vizard Agent actually do?

Here is the order on a real talking-head edit where several phrases were re-voiced in a clone of the speaker's own voice. The cutaways went on early and were still adjusted twice, which is the normal shape of this job.

1. Transcribes the original with word-level timings.
2. Assembles the corrected speech track from the clone.
3. Level-matches the new lines to your natural speech.
4. Checks the joins so no level jump gives them away.
5. Sources b-roll that suits what each line is about.
6. Lays the cutaways across the replaced passages.
7. Renders and pulls review frames.
8. Verifies the b-roll fully covers each re-voiced line.
9. Widens the cutaways where the cover falls short.
10. Re-renders and tightens the transitions at the joins.

Step eight is the whole reason this article exists. A cutaway placed from the timings usually looks right and is frequently a few frames short at one end, which is precisely where a viewer's eye lands as the picture changes.

Step three is what stops the ear giving it away instead of the eye. A cloned line at a different level from the surrounding speech is as obvious as a mismatched mouth, so Vizard Agent measures your natural speech level and matches to it.

Step five is where the cover stops being an excuse. A cutaway chosen because it happens to be to hand reads as an escape from the speaker, so Vizard Agent sources each one against what that particular line is talking about and the cover doubles as illustration.

Step ten matters because widened cutaways start to feel like an edit in themselves. Once the coverage is right, Vizard Agent tightens the transitions so the video does not read as a sequence of escapes from the speaker's face.

What does the result look like?

A talking-head video where your voice says the corrected words and the picture is somewhere else at each of those moments. Nobody watching it can tell you which of the lines were changed, because there is nothing on screen at those points that disagrees with the audio.

Vizard Agent can tell you, though, which is exactly what you want when a fifth correction arrives next week.

When does this not work well?

Cutting away is not always available. If the shot is the point, or the replaced passage runs for a long stretch, covering it turns into hiding the speaker for a chunk of the video, which viewers read as an edit even without knowing why.

• Long replaced passages. The cover becomes the video.
• Nothing to cut away to. No relevant b-roll exists.
• A shot that must stay on the face. A direct address.
• Back-to-back corrections. They merge into one long escape.
• Very short clips. There is no room for a cutaway.

How do you fix a result that came back wrong?

Point at the moment. Because Vizard Agent knows which passages were replaced, "I can see my mouth at 0:32" resolves to a specific cutaway and a specific end of it rather than to a general note about b-roll.

• "I can see my mouth at the end." The cutaway gets extended.
• "Too much b-roll now." Ask for lip-sync on one instead.
• "The b-roll does not match." Vizard Agent re-sources that one.
• "The level jumps." Vizard Agent re-matches that line.

How does Vizard Agent compare to doing it yourself?

By hand the cutaways go on by eye against a waveform and are usually a little short, because the natural instinct is to reveal the speaker as soon as the audio allows. The verification pass is the part nobody does twice.

Common questions

Why not just lip-sync it? You can, and Vizard Agent will. Cutaways are cheaper and safer.

Will the b-roll be relevant? Yes. Vizard Agent sources it from what the line is about.

How much extra cover is needed? Enough to land before and after. Vizard Agent verifies rather than guesses.

Can I mix both approaches? Yes. Lip-sync the close-ups, cut away on the rest.

Does it match the new audio's level? Yes, to your own natural speech level.

What if I have no b-roll? Vizard Agent can find or generate it. Say which you prefer.

Will it tell me which lines changed? Yes, with timings, which matters for the next round.

Can it cover a whole paragraph? It can, but Vizard Agent will warn you it starts to show.

Does this work with a cloned voice? Yes. That is exactly the case it is for.

Will the cutaways feel abrupt? Not once Vizard Agent tightens the transitions.

Can I choose the b-roll myself? Yes. Hand Vizard Agent the clips and it places them.

What about the captions? They follow the corrected audio, not the original.

Does it need the original take? Yes, the picture comes from it.

Why does it fail at the end and not the start? Because a cutaway is usually cued to the first word, not the last.