Vizard Agent

How to set the mix yourself when the music is fighting the voice

Last updated 2026-09-04 · 7 min read

Tell Vizard Agent which layer to move and by how much — "music down eighty per cent, sound effects down fifty, voice up fifty". It re-mixes from the stems it already has, so nothing is generated again, and the picture, the voice performance and the timing are all untouched.

What is the short version?

A mix note is not a creative note. "The music is too loud" invites a rethink; "the music down eighty per cent" is an instruction that can be carried out exactly, and it costs almost nothing because nothing is remade.

  1. Go to Vizard Agent with the current version.
  2. Name the layer and the amount — a percentage is fine.
  3. Say it is a mix change, not a regeneration.

What do you need before you start?

The delivered video and an opinion. You do not need to know decibels — relative amounts are a perfectly good language for this, and they are how everyone who is not an audio engineer actually thinks about a mix.

What do you type into Vizard Agent?

Say all the layers you want moved in one message. Mix changes interact with each other — lifting the voice and dropping the music are two different ways of making the same change, and asking Vizard Agent for them in separate rounds reliably means overshooting one of them.

Prompt

Lower all the sound effects by 50%, lower the backing track by 80%, and increase the voiceover by 50%. Don't regenerate anything — this is just the mix.

Variants worth knowing:

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real product ad whose music bed was sitting on top of the narration. The mix pass at the end is short precisely because every asset it needs already exists and nothing has to be made a second time.

  1. Checks the voiceover parameters used in the original build.
  2. Generates the narration and the music bed for the first version.
  3. Transcribes the narration for word-level timings.
  4. Re-generates the narration where the price was mispronounced.
  5. Re-transcribes the corrected version to keep the timings true.
  6. Mixes the voiceover, the music bed and the sound effects together.
  7. Applies the requested level changes to the existing stems.
  8. Uploads the version with the adjusted mix.

Step seven is the whole of it. Because the voice, the music and the effects exist as separate stems, changing the balance is arithmetic on files that are already made — no voice is re-recorded, no track is re-searched, and the word timings that the captions depend on do not move.

Step four is worth noting as the contrast. That was a real regeneration, because the narration said a price wrong and no amount of mixing fixes a wrong word. Knowing which of the two kinds of note you are giving is most of the skill here.

What does the result look like?

The same video, audibly different. The narration now sits in front, the music supports it from underneath, and the effects punctuate rather than compete with it — while the picture, the cut, the captions and every timing in the file are bit-for-bit what you already approved.

Turnaround is short. A mix note is the cheapest kind of change there is, which is worth knowing before you describe a level problem as though it were a creative one.

When does this not work well?

Mixing only moves the balance between things that already exist in the file. It cannot fix a recording that was captured badly in the first place, and no amount of level adjustment by Vizard Agent rescues a music track that was simply the wrong choice for the video.

How do you fix a result that came back wrong?

Give another relative amount. Vizard Agent keeps the stems and the last mix, so a second adjustment moves from where you are rather than from the original — which is why "another twenty per cent down" is a usable instruction.

How does Vizard Agent compare to doing it yourself?

If you have the stems and a mixer, this is five minutes. The problem is that most delivered videos arrive as one flattened file, at which point separating the voice from the music is a much harder job than setting the balance would have been.

By hand Vizard Agent
Access to the layers Only if you kept the stems The stems are still there
Making the change Faders A sentence with amounts
Risk of a rebuild You would not Say "don't regenerate"
Captions and timing Unaffected Unaffected
Matching another video By ear Measured against the file

Common questions

Do I need to say decibels? No. Vizard Agent takes percentages or plain words like "much quieter".

Will it regenerate the voice? Not for a mix note, if you say so. Vizard Agent works from the existing stems.

Can it duck the music under speech? Yes, and Vizard Agent usually recommends that over lowering it throughout.

Does the video re-render? Vizard Agent remixes and remuxes the audio. The picture is untouched.

Can I ask for a loudness target? Yes. Vizard Agent measures and delivers to a stated target.

What if there are no stems? Vizard Agent will say so, and separating a flat mix is a different job.

Can it match another video's balance? Yes. Vizard Agent measures that file and matches it.

Does it cost much? Very little. Vizard Agent generates nothing again.

Can I change three layers at once? Yes, and you should. Vizard Agent applies them together.

Will the captions still line up? Yes. A mix change moves no timing at all.

Can I hear a version before you commit? Yes. Ask Vizard Agent for a short sample of the new balance first.

What about the music at the very end? Say so separately — a closing swell is often worth lifting rather than lowering.

What if the voice is distorted? That is in the recording. Vizard Agent can only re-record, not re-mix, out of it.

Can it make the music more energetic instead? That is a new track, not a mix note — Vizard Agent will treat it as such.

Does it check the result? Yes. Vizard Agent measures the new mix before uploading.

Why separate mix notes from creative notes? Because one is arithmetic on files that exist and the other is a rebuild, and they cost very different amounts.