Feature
Custom song from lyrics: the words decide more than the genre
Paste your lyrics, describe the sound, pick a length up to 120 seconds, and a sung track comes back for 20 credits. The description picks the genre. The words decide whether anybody can follow it. That is the part most people get wrong. This page covers the two boxes, the word budget, how to write for half attention, and when to leave the lyrics box empty instead.

- One sung track, any length
- 20 CR
- Longest track available
- 120s
- Attempts from 300 free credits, no card
- 15
The two boxes and what each one controls
The lyrics box takes the words to be sung. Whatever you paste, somebody sings.
The description box takes the sound: genre, mood, instruments, tempo, and whether the lead vocal is male or female.
Structure markers in square brackets, like verse and chorus, help the arrangement understand what is repeating.
Pick 15, 30, 60, 90 or 120 seconds. The price is 20 credits at any of them.
The 300 free credits with no card cover fifteen attempts, which matters here because lyrics rarely land on the first return.
There is no editing after generation. If a line lands badly you change the words or the description and spend another 20 credits.
How many words fit in each length
Roughly sixty singable words a minute. Paste more than that and lines get rushed or dropped.
- 15 secondsAbout 15 words
- 30 secondsAbout 30 words
- 60 secondsAbout 60 words
- 120 secondsAbout 120 words
Sung words are slower than spoken words, and that is the constraint
A sung line carries roughly a third of the words a spoken line does in the same time.
So sixty seconds is about sixty singable words, not the hundred and fifty a voice-over would fit.
Paste a paragraph into a thirty second track and the result rushes, drops lines or repeats the first half.
Count your words before you generate. That check costs nothing and it prevents the most common disappointment in this module.
Short lines with repeated ends sit better than long ones, because the arrangement has somewhere to breathe.
If the lyrics matter more than the music, write fewer of them rather than asking for a faster tempo.
A faster tempo does not buy words. It buys a rushed delivery of the same words, and a rushed sung line is the hardest thing on the page to follow.
This is the whole editor
Highlight a phrase and a clip lands on those exact words. No timeline, no keyframes, no layers.
Write for a listener who is half paying attention
In an ad, nobody is studying the lyrics. They hear a phrase, and either it means something immediately or it is texture.
So put the one sentence you want remembered at the start and repeat it at the end.
Avoid internal clauses and clever inversions. A sung line that has to be unpacked is a line that will not be.
Concrete nouns survive singing. Abstract nouns disappear into the melody.
Repeating a short phrase is not lazy writing here, it is the only mechanism the format has for making something stick.
Read the words aloud at singing speed before you paste them, and cut anything you stumble on.
Stumbling is the reliable signal. A line that trips your own tongue will trip a sung delivery, and there is no way to fix it once the track returns.
- The key line first, repeated last
- Concrete nouns over abstract ones
- No internal clauses, no inversions
- About sixty singable words a minute
A sung track and a voice-over cannot share an ad
Two sets of words at once means the listener follows neither, because attention for language is single-channel.
So a custom song belongs in ads with no voice-over, where the words on the track are the words of the ad.
If there is a voice-over, generate an instrumental for that section by leaving the lyrics box empty, at another 20 credits.
The editor gives you turn music on, pick the track, and set how loud it sits. There is no ducking and no stem separation.
That means the separation has to be structural. The sung part owns its own seconds and the voice owns the rest.
Check the result on a phone at half volume. A sung vocal and a spoken one collide hardest on a small speaker.
Deciding whether your ad wants a song at all
Three questions, and most ads answer them in under a minute.
Is there a voice-over?
Yes: instrumental, not a song
Under 60 words?
No: cut before generating
One line to remember?
No: a bed will do the job
Generate
20 CR, then listen on a phone
Save the description, and know the four limits around it
One habit is worth adopting from the first attempt. Save every description that produced something usable, alongside the words you fed it.
Regenerating from the same description gives you a sibling track rather than a stranger. That is how a small brand builds a consistent sound.
No fixed piece has to be committed to, and the words can change every week. For somebody shipping ads weekly, that is the practical version of a sonic identity.
Four plain limits sit around it. A full song for release is the wrong ask. 120 seconds is a hard ceiling and there is no stem export.
There is no editing a vocal line, retiming a section or fixing one word. You regenerate for 20 credits.
A specific recognisable song or an existing melody is a rights negotiation.
And lyrics carrying legal or medical claims are harder to hear accurately when sung, which is a compliance question rather than a creative one.
The track is generated for you rather than licensed from a catalogue. Confirm the usage terms that apply to your own account before running it behind paid spend.
Keep the lyrics, the description, the date and the file together. That record answers a question asked much later.
Questions people ask
- How many words fit in a 30 second track?
- About thirty singable words. Sung delivery carries roughly a third of what speech does in the same time, so count before you paste. Overlong lyrics are the most common reason a return sounds rushed.
- Can I choose who sings it?
- You can ask for male or female lead vocals in the description, along with the genre, tempo and instruments. It is a description rather than a casting choice, so expect variation between attempts on the same brief.
- Can I edit the track after it comes back?
- No. There is no retiming, no stem separation and no way to fix one line. You adjust the words or the description and generate again for 20 credits, which is how the module expects to be used.
- Who should not use this?
- Anybody making an ad with a voice-over, because sung and spoken words cannot share an ad. Generate an instrumental instead by leaving the lyrics box empty. Also anybody wanting a releasable song, since 120 seconds and no stems is the wrong shape for that.
Count the words, put the key line first, and leave the lyrics box empty whenever somebody is talking.