How to add background music to an AI video without burying the dialogue

Two decisions, one before the render and one after

Most guides on this start in the editor: drop the track in, pull the slider down, let the tool duck it under the voice. Invideo, Pictory and Tella all have a version of that, and some do the ducking for you.

With AI video the first decision comes earlier. Several engines now make their own sound in the same pass as the picture, and on some of them that sound costs extra. So before you pick a song, you decide whether each shot renders silent or with sound, and that choice changes the bill. The second decision is the level, and our editor has no automatic ducking, so this page shows how we set it by hand and what the numbers on the gain control mean.

Decision 1: silent or voiced render

engine	720p silent, per second	720p with sound, per second
Veo 3.1 Fast	20	30
Kling O3 Pro	24	30
Seedance 2.0	45	45
Seedance 2.0 Fast	50	50
Seedance 2.5	70	70

On Veo 3.1 Fast and Kling O3 Pro, sound is a switch with a price. On every Seedance model the clip comes with sound and the rate is the same either way.

That gives three cases:

A montage under music, nobody speaking. Render silent on Kling O3 Pro or Veo 3.1 Fast. The music covers everything, so paying for ambience you will turn down is waste.

A character speaks over the music. Render that shot with sound, since nothing else can move the mouth, and keep the other shots silent.

Seedance. The sound is free, so keep it, and write "no music" in the sound line. Generated sound can include a music-like bed, and one Reddit thread on the subject found no reliable way to strip music out of a generated clip afterwards. Asking for none in the prompt is cheaper than fixing it. Our sound effects page covers how to write the sound line.

A 30-second music-led reel, priced

Six shots, one with a spoken line, at 720p:

engine	all silent	one voiced shot of 5 s, rest silent	all with sound
Kling O3 Pro (six 5 s shots)	720	750	900
Veo 3.1 Fast (8 + 8 + 8 + 6 s)	600	680	900
Seedance 2.0 (six 5 s shots)	1,350	1,350	1,350
Seedance 2.5 (six 5 s shots)	2,100	2,100	2,100

Veo serves only 4, 6 or 8 seconds, so its version is four shots, and the voiced one is an 8-second shot: 80 credits more than silent. With a character reference attached, every Veo shot is 8 seconds, which would make the reel 32 seconds. For a music-led reel with a recurring character we build on Kling: 750 credits, with one line spoken and the music carrying the rest.

Decision 2: putting the music in

In the studio's editor, music goes on the audio lane under the clips.

• Upload. MP3, WAV or M4A, up to 100 MB. Uploading costs no credits.
• Place it. Drag the block along the lane to where it starts. Drag its left edge to trim the start of the song, its right edge to trim the end.
• Set the level. Each piece has a gain from 0 to 1, in steps of 0.05. One is the file as it is; the lane only turns music down.
• Up to 20 pieces. Enough for a bed and a handful of stings, or for splitting one song into sections.

Two things about the mix are worth knowing before you export. Adding music does not turn your clips' own sound down. The mix leaves the dialogue at the level it was rendered at and moves only the music, with a limiter at the end that catches the sum if it gets too loud. And the export is exactly as long as the video, so a three-minute song under a 30-second reel stops when the picture does.

What the gain numbers mean

Mixing guides talk in decibels. The gain control is a multiplier on the music file's own level, and the conversion is fixed:

gain	change in level
1.0	0 dB, unchanged
0.5	about -6 dB
0.25	about -12 dB
0.15	about -16.5 dB
0.1	-20 dB
0.05	about -26 dB

The guides disagree on the target. One says keep music 10 to 15 dB under the dialogue; others give -18 to -24 dB for short-form with a voice over it. Both are measured against the voice, and a commercial song is usually mastered loud, so the gain that works depends on the track. We start at 0.15 under speech and 0.5 where nobody talks, then listen on a phone speaker before exporting.

Dipping the music under a line, by hand

The lane does not duck automatically and does not fade; when a piece ends, it ends. So for a reel where the music should sit loud, drop under a spoken line, and come back, we cut the song into pieces:

1. Add the song and trim it to end where the line starts. Gain 0.5.
2. Add the same upload again, trim its start to pick up where the first piece ended, and its end to where the line finishes. Gain 0.15.
3. Add it a third time for the rest of the reel. Gain 0.5.

Because each piece starts where the last stopped in the song, the music plays straight through and only the level steps. A step is less smooth than a fade, so we cut the dips at a beat or under the first word of the line, where a change in level is expected. For an ending, two or three short pieces stepping down (0.5, 0.25, 0.1) do the job of a fade-out.

Exporting

An export out of the editor is 20 credits on the free, Lite and Starter plans and free from the Creator plan up. Captions are 8 credits per started minute on the same entry plans and free from Creator up. The editor exports up to 300 seconds of output.

Use music you have the rights to use. That is between you, the track's licence and the platform you post on, and nothing in the editor checks it.

To try the dip on your own reel, render one voiced shot and the rest silent, then open the editor in the studio and lay the song in three pieces. For a narrated version of the same idea, see narration over silent clips.

FAQ

How do I add background music to an AI generated video?

Upload the song to an editor's audio lane under the clips, trim it to the video, and turn it down so speech stays clear. In the ReelDouble editor, MP3, WAV and M4A files up to 100 MB upload free, and each piece has a gain from 0 to 1.

How loud should background music be under dialogue?

Guides suggest 10 to 24 dB below the voice, and the right level depends on how loud the track was mastered. On a 0 to 1 gain control, 0.15 is about 16.5 dB down. Start there under speech and check on a phone speaker.

Should I render AI video with sound if I am adding music?

Render silent where music covers the shot and nobody speaks, since Kling O3 Pro and Veo 3.1 Fast charge less silent. Seedance always includes sound at the same rate, so keep it and write no music in the sound line.

How do I lower music only while a character talks?

If your editor has no automatic ducking, split the song into pieces on the audio lane: normal level before the line, lower during it, normal after. On ReelDouble you add the same upload again and trim each piece so the song plays straight through.

Can I remove music an AI video generator added to my clip?

There is no reliable way to strip music out of a finished generated clip. Ask for no music in the prompt's sound line, or render silent on an engine that offers it, and add your own track in the edit.