AI music for Reels: the in-app library is not for paid ads

You type how it should sound. Warm acoustic guitar, organic, gentle. Modern hip-hop beat, crisp drums, deep bass. Whatever you actually hear.

Suggestion chips write into that description rather than replacing it. Start from a genre, then add tempo, mood and instruments.

The lyrics box decides whether anybody sings. Empty means instrumental, and anything typed in gets performed.

Choose 15, 30, 60, 90 or 120 seconds. The price is flat at 20 credits whichever you pick.

The 300 free credits with no card cover fifteen tracks, which is enough to work out what your brand sounds like.

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

Instagram surfaces a large library of commercial tracks inside the app, and creators use it constantly.

The permission attached to that library is normally scoped to organic, non-commercial posting. It commonly does not extend to paid promotion or to business accounts in the same way.

That is the actual reason a marketer ends up looking for generated music, and it has nothing to do with the catalogue sounding bad.

A generated track is made for you rather than licensed, so no other advertiser has the same recording and there is nothing for a matching system to compare it against.

That removes a common failure mode. It is not the same as being told your usage is safe, and nobody honest will tell you that.

Confirm the terms that apply to your own account. Save the description, the date and the file alongside the ad.

A phone speaker reproduces almost nothing below roughly 500 Hz, and most viewers never reach for headphones.

So the deep, warm low end that made the track feel expensive on your monitor is not present at the point of contact.

What is present is the midrange, which is where a speaking voice lives, so a full-sounding track competes with your own words.

Brief for space rather than for size. Sparse, high percussion, almost nothing in the middle beats anything containing the word big.

Check the finished ad on a phone held at arm's length with the volume at half, which is the realistic listening condition.

If the voice-over stops being clear at that point, the track is too busy in the middle and the fix is a new description, not a lower level.

A sung lyric and a spoken voice-over cannot share an ad. A listener cannot follow two sets of words at once.

So the working pattern is a sung opening for the first two seconds and an instrumental bed under everything after it.

That is two generations at 20 credits each, which is cheaper than any amount of fiddling with a single track's level.

In the editor you turn music on, pick which track, and set how loud it sits. There is no ducking and no stem separation to rescue a bad pairing.

If the ad has no voice-over at all, a sung track can carry the whole runtime and often outperforms an instrumental.

Decide which of those two shapes your ad is before you generate anything, because the description you write is different for each.

Budget three attempts on a sound you have never made. Budget one on a sound you are repeating from last month.

That habit keeps the music line on a campaign small enough that nobody has to think about it.

Two ads are stronger with nothing underneath. A dense voice-over with no pauses gains nothing from a bed and loses clarity to it.

A demonstration where the product makes a sound worth hearing is better with the real audio.

Test the ad muted first. If it works without music, adding music is a decision rather than a requirement.

Four plain limits are worth knowing before you start. There is no way to edit a track once it returns. No retiming, no stems, no trimming.

When a track is close but wrong, adjust the description and spend another 20 credits. That is how the module is designed to be used.

The longest single track is 120 seconds, so a longer piece has to be assembled elsewhere.

Stems, tempo control and automated ducking do not ship here. A full audio editor is the right purchase for those.

And a specific recognisable song is a licence negotiation rather than a generation problem.