How to get a clean version of your video with no captions or music

Last updated 2026-09-05 · 8 min read

What is the short version?

A clean version is the same edit with layers switched off, not a new edit. Asking for it after delivery is normal and cheap, provided it is built from the same assembly rather than reconstructed from the export.

1. Go to Vizard Agent with the delivered video.
2. Ask for the same video with no captions and no music.
3. Ask it to confirm both actually came off.

What do you need before you start?

Nothing except the request, and a sentence about which layers you want gone. It is worth being explicit — captions and music are usually meant, but sound effects, lower thirds and end cards are separate things and each one has to be named.

• The delivered video, or the project it came from.
• Which layers come off. Captions, music, graphics.
• What stays. Speech, effects, the cut itself.
• What it is for. Dubbing, licensing, a re-edit.
• The format you need it in. Usually the same.

What do you type into Vizard Agent?

Ask for "exactly the same video" and then list what comes off. Those five words matter, because a request for a version without captions can otherwise be read as a licence to improve the pacing now that the words are gone.

Prompt

Variants worth knowing:

• "Exactly the same video." Nothing else changes.
• "Without captions and without music." Name each layer.
• "Check it actually came out." Quiet is not gone.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on real deliveries that needed clean versions alongside the finished ones. Half the list is verification, because "the music is off" is a claim that can be checked rather than asserted.

1. Renders the same build without the caption layer.
2. Checks the clean version on frames from the render.
3. Removes the background music and measures the audio.
4. Measures the audio levels in both versions.
5. Finds a section with no speech and measures its overall loudness.
6. Compares the difference between the two versions at that point.
7. Removes a specific sound you flagged and verifies it is gone.
8. Compares both soundtracks at the exact moment it should have left.
9. Uploads the clean version alongside the delivered one.

Step five is the check that actually proves it. Over speech, a music bed sitting low is almost impossible to hear the absence of, so the measurement is taken where nobody is talking — there the difference between the two versions should be silence and anything else is a bed that survived.

Step seven is worth knowing you can ask for. A clean version and a specific removal are the same kind of request — one sound at one moment can be taken out and verified the same way the whole music bed can, without touching anything else in the mix.

Step one is the part that protects the edit. Rendering from the same assembly with a layer switched off guarantees the cut, the timing and the mix of everything else are identical, whereas rebuilding from the export invites small differences you will find later.

What does the result look like?

Two files that are the same video: the delivered one and a clean one with nothing burnt in and no bed underneath. The speech, the effects, the cut and the length are identical, and the difference between the two soundtracks measures as the music you took out.

You can dub it, license it, or hand it to somebody else to finish.

When does this not work well?

A clean version can only remove what was added as a layer. Anything baked into your source footage — a caption burnt in before you sent it, music recorded in the room — is part of the picture or the recording rather than something that can be switched off.

• Burnt-in captions from the source. Those need covering.
• Music recorded in the room. It is part of the location sound.
• Only the export available. Reconstruction is imperfect.
• Effects tied to the music. Cuts on the beat still look like it.
• A mixed-down soundtrack. Separation is harder than switching off.

How do you fix a result that came back wrong?

Say what you can still see or hear. Vizard Agent keeps the measurements of both versions, so "there is still music under it" resolves to a measured difference rather than an argument about what somebody thinks they heard.

• "There is still a bed underneath." Measured and re-rendered.
• "A caption survived." Located and the layer switched off.
• "The timing is different." Rebuilt from the same assembly.
• "I need the effects gone too." Named as a separate layer.

How does Vizard Agent compare to doing it yourself?

Without the project, a clean version means removing music from a finished mix, which is separation rather than subtraction and always leaves something behind. With the project it is trivial, and the only real question is whether anybody checks the result.

Common questions

Is this a new edit? No. Vizard Agent renders the same build without a layer.

Can I have captions off but music on? Yes. Tell Vizard Agent which layers to drop.

What about the sound effects? They stay unless you name them. Tell Vizard Agent if they should go.

Can I get the subtitle file separately? Yes. Ask Vizard Agent for it alongside the clean version.

Will the length be the same? Yes. Vizard Agent keeps the cut and the timing exactly.

Can it prove the music is gone? Yes. Vizard Agent measures both versions where nobody is speaking.

What if the music is in my source clip? Then it is location sound, and Vizard Agent will tell you so.

Can I get the music on its own? Yes. Vizard Agent exports it as its own file for reuse.

Do I need this for dubbing? Usually yes, and Vizard Agent finds captions off matters most there.

Can it do this months later? Yes, if the project or the source is still available.

Will graphics come off too? Only if you ask. Vizard Agent treats each layer separately.

What about the end card? Tell Vizard Agent whether it stays; it is a separate decision.

Can I get both formats clean? Yes. Vizard Agent renders the clean pass in each one.

Why not just strip the delivered file? Because separating music out of a finished mix leaves traces, and never including it does not.