Instrumental for a voice-over: the empty box is the whole trick

Leave the lyrics box empty. That is the only thing separating an instrumental from a song here.

Type the sound in the description box in your own words, comma separated. Ambient pads and texture, almost no rhythm is a complete brief.

Pick 15, 30, 60, 90 or 120 seconds. Match it to the ad rather than generating long and trimming.

Price is 20 credits flat at every length. The 300 free credits with no card cover fifteen attempts.

In the ad editor the music controls are turn it on, pick the track, and set how loud it sits.

There is no ducking, no stems and no tempo slider, so the description carries all of the control you have.

A speaking voice occupies roughly 500 Hz to 4 kHz, and the consonants that carry meaning sit at the top of that band.

Anything in the track living in the same range masks those consonants, and the listener stops following without knowing why.

So a good brief names absences. Sparse. No pads. Nothing in the midrange. Low end and high percussion only. No vocals.

It reads oddly and it works. A brief asking for warmth, fullness or energy returns exactly the midrange you cannot afford.

Ask for slow, around 70 BPM, when the voice-over is fast, because two competing rhythms make both feel hurried.

The word to avoid is big. It reliably returns a wall of exactly the frequencies your voice needs.

When a bed is fighting the voice, the instinct is to pull the music down until the problem goes away.

That works until the music is so quiet it contributes nothing, at which point you paid 20 credits for silence.

The actual problem is that the track is busy in the same band as the speech, and no level solves an overlap.

Generate a sparser track for another 20 credits. That fixes it in one step and leaves the music audible.

Judge it on a phone at half volume held at arm's length, not on headphones, because headphones separate what a phone cannot.

If the words are clear on a phone and the music is still noticeable, the description was right.

A fast, urgent read wants a slow, spacious bed. A slow, calm read can carry more movement underneath it.

That is the opposite of what most people brief, because they describe the brand rather than the delivery.

A bed is a supporting decision. It should be described against the voice-over that already exists, which means recording the voice first.

If you have the take, listen to it, then write the description while it plays in your head.

That order costs nothing and it produces a usable bed in one or two attempts rather than five.

Save the descriptions that worked. Three good sentences cover most of what a brand ever needs behind speech.

Reusing a description also gives you a family of beds rather than a set of unrelated ones, which is how a run of ads starts sounding like it came from the same place.

Three saved descriptions and a phone to check on will cover a year of ads. Nobody has to open an audio editor once.

What this does is narrow. It turns a sentence about a sound into a usable bed in one step, cheaply enough that a wrong answer costs almost nothing.

That converts a production problem into a writing problem. For anybody whose job is making ads rather than mixing them, that is the better problem.

Two ads want no bed at all. A dense voice-over with no pauses gains atmosphere and loses clarity, which is a bad trade on a direct response ad.

A demonstration where the product makes a real sound is stronger with that sound than with music over it.

Play the ad muted, then with voice only. If voice only holds, the bed is optional rather than required.

Four plain limits sit around all of this. No ducking, no stems, no editing after generation, and 120 seconds is the ceiling.

Frame-accurate audio work and automated ducking belong to a full audio editor. This module is not competing for that job.

The track is generated for you rather than licensed from a catalogue. Confirm the usage terms that apply to your account before running it behind paid spend.