How to stop a multicam edit cutting to the wrong camera

Last updated 2026-09-04 · 7 min read

What is the short version?

Recorded meetings and webinars come with an automatic camera switcher that reacts to sound. It cuts to whoever coughed, and it cuts away from whoever is still talking — and a clip pulled from that recording inherits both mistakes.

1. Go to Vizard Agent with the recording or the clips.
2. Say some cut to the wrong person or end mid-answer.
3. Say to scan every clip, not just the ones you spotted.

What do you need before you start?

The source recording and a note of what you noticed. You do not have to find every instance yourself — one example of each fault is enough for Vizard Agent to know what it is looking for across the rest.

• The recording. However long.
• One example. Of a wrong cutaway or a cut-off ending.
• Who should be on screen. When, if it is not obvious.
• The clip length. Whether endings may run over.
• Any fixed openings. A hook that must survive.

What do you type into Vizard Agent?

Report the fault type rather than the timecode. A timecode gets that clip fixed; naming the fault gets every instance of it found, and in a recording this long there are always more than the two you noticed.

Prompt

Variants worth knowing:

• "Find them all." Turns one report into a scan.
• "Endings may run over." Permission to extend past the target.
• "Keep the hooks." Protects openings that already work.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on twelve clips pulled from a recorded webinar nearly seven hours long. The repair pass proper starts at step five, after the whole set had already been rendered, reviewed and delivered once.

1. Checks which minutes are full-frame speaker in the source.
2. Locates the speaker's face and checks each planned clip's frames.
3. Renders the clips with captions and graphics.
4. Reviews the rendered clips and repositions the graphics.
5. Watches the riskiest clips back and measures the mix.
6. Reads the watch reports and the loudness numbers together.
7. Locates the bad cutaways and the missing endings.
8. Pinpoints the cutaways and finds the answer one clip never reached.
9. Renders the repaired clips and checks them.
10. Re-times the captions on a clip whose ending changed.
11. Confirms that ending now completes.
12. Watches the remaining eight clips, scanning for wrong-camera shots.
13. Rebuilds those clips with fixed openings and endings.
14. Renders and uploads the rebuilt set, then checks them.

Step twelve is the one that matters. Having found the fault in four clips, it went and watched the other eight looking specifically for the same thing — because an auto-switcher that failed in one part of a seven-hour recording failed in others too, and the clips you did not happen to watch closely are exactly where it hides.

Step ten is the consequence nobody plans for. Extending a clip so the answer completes moves its end, which means the captions no longer run to the right place, so they are re-derived rather than left to run short.

What does the result look like?

Clips where the camera is on the person actually speaking, and where every answer runs to its end. Some come back slightly longer than they were, because completing a thought takes however long it takes and Vizard Agent will not trim it short to hit a number.

The scan is worth as much as the repair. Knowing that all twelve were checked — rather than the four you happened to notice — is what makes the set safe to publish without watching every one yourself.

When does this not work well?

The fix depends entirely on the right shot existing somewhere in the recording. Where the automatic switcher made the only camera decision and no wide shot or alternative angle was ever recorded alongside it, Vizard Agent has nothing to cut to instead and will say so.

• A single recorded angle. Nothing to switch to.
• Everyone talking at once. No correct camera exists.
• A hard clip length. Completing an answer needs the seconds.
• Screen-share sections. Nobody is on camera at all.
• Poor source layout. Some recordings bake the switch into the file.

How do you fix a result that came back wrong?

Say which clip you mean and which of the two faults it has. Vizard Agent keeps the transcripts, the per-clip layout analysis and the watch reports from the scan, so repairing one clip leaves the others exactly as they were delivered.

• "This one still cuts away." Re-scanned and re-cut on that span.
• "The ending is too long now." Trimmed to the end of the thought.
• "Wrong person the whole way through." Re-cut from the right angle.
• "Captions run short." Re-derived from the extended cut.

How does Vizard Agent compare to doing it yourself?

By hand this means watching twelve clips properly, which sounds easy and is the reason it does not happen. Clips get spot-checked at the start, published, and the wrong-camera moment in the middle of clip nine is found by a viewer.

Common questions

Can it work from the raw recording? Yes. Vizard Agent locates the speakers itself and cuts from there.

What if there is only one camera? Then a wrong cutaway is not fixable. Vizard Agent will tell you.

Will clips get longer? Some will. Vizard Agent spends the few seconds an answer needs.

Can I cap the length? Yes, though Vizard Agent will say when an ending will not fit.

Does it check the ones I did not report? Yes — that is the point of asking it to scan them all.

What about screen-share sections? Vizard Agent treats them separately; there is no camera to be wrong.

Will the captions still match? Yes. Vizard Agent re-derives them whenever an ending moves.

Can it tell me which clips it changed? Yes. Ask Vizard Agent for the list with the rebuilt set.

Does it re-render everything? No. Vizard Agent re-renders only the clips that needed repair.

Can it fix the opening too? Yes. Vizard Agent rebuilds openings and endings in the same pass.

What if two people talk over each other? There is no correct camera. Vizard Agent flags it rather than guessing.

How long can the source be? Long. Vizard Agent handled nearly seven hours here.

Can it keep a hook I already like? Yes. Name it and Vizard Agent leaves that opening alone.

Why does the switcher get it wrong? Because it is reacting to sound rather than to meaning, and a cough is louder than the end of a sentence.