No. A person reads your story, writes the lyrics around the details you give us, produces the song and cuts it to your footage. We use software in the studio the same way any producer does. Every song is checked and finished by a human being before it reaches you, which is why it takes a couple of days rather than a couple of seconds.

How long does it take?

24 hours from the moment we get your brief, on every package. If you are closer to the date than that, write to us before you order and we will tell you honestly whether we can make it.

What do you actually need from me?

A few sentences about the people the song is for: how they met, what they call each other, one or two things only their friends would know. Then one photo of them, straight off your camera roll. The written details do most of the work. A song about nobody in particular is the one thing we will not send you.

Can I have it in Hindi?

English, Hindi, or Hinglish mixed the way people actually speak it. Hinglish is the most requested and usually the best of the three, because it lets the song sit in both worlds at once.

What if I do not like it?

Every package includes revisions: one on Single, two on Signature, three on the Wedding Pack. Tell us what is not landing and we will rework it. In practice most changes are small, a name pronounced differently or a verse that needs to sit lower.

You do, for personal use: play it at your wedding, post it, keep it forever. We may ask separately whether we can show it as an example, and we will never do that without you saying yes in writing first.

Do I need permission to upload the photo?

Yes, and we ask you to confirm it. Everyone recognisable in the photo needs to be happy for you to send it to us for this purpose. We store it privately, serve it over expiring links, and delete it on request.

Can you write for something sad?

Yes. Memorials and songs for someone who is gone are some of the most meaningful work we do. Write to us first rather than ordering through the site, and we will handle it with the care it needs.

How an AI song is actually made, start to finish

Not the marketing version. The actual file structure, the actual prompts, and the four hours of unglamorous work that sit between a good idea and something you would play to somebody.

People assume this is one box you type a sentence into. It is closer to four separate jobs, and the interesting parts are the joins between them. This is the process I use for every song that goes out, written down properly.

One: the brief, which is the whole thing

Before any tool is opened, I need facts. Not adjectives. The difference between a song that works and one that does not is almost entirely decided here.

Useless: they are very much in love, she is beautiful, they have been together a long time.

Useful: they met at a friend's birthday in Leeds in 2019, he spent the entire night pretending he could dance, she let him carry on, and he still calls her Doctor Sahiba because she corrects everybody, including him, especially him.

The second one contains a chorus. The first one contains nothing. Half of what our order form does is drag people from the first kind of answer to the second, which is why it asks for the story rather than for a list of qualities.

Two: writing the lyric

I draft with a language model, then rewrite by hand. The draft is useful for structure and useless for meaning. Its instinct is to reach for the nearest available image, which is always the one every other song has used.

The rule I work to: every verse must contain one thing that could not be said about any other couple. If a line would survive being moved into a different song, it comes out.

Sections get labelled explicitly, because the music model reads them:

[Intro], kept short, usually instrumental
[Verse 1], the situation, concrete and specific
[Pre-Chorus], the turn, where the feeling arrives
[Chorus], the title made literal, repeated exactly
[Verse 2], the complication or the passage of time
[Bridge], genuinely different, or the final chorus lands flat
[Final Chorus], often with one word changed

Three: the style prompt

This is the part with the most folklore around it and the least documentation. A style prompt is one comma-separated line describing the record you want, and it behaves less like an instruction and more like a search query across everything the model has heard.

What works, in order of how much it matters:

1Instruments, named specifically. Harmonium and tabla, not traditional Indian instruments.
2Vocal character. Warm male baritone, restrained, close-mic'd.
3Era and production. Nineties playback, wide reverb, strings slightly forward.
4Tempo and feel. Slow six-eight, unhurried.
5Mood, last and briefly. Everything before this has already done the work.

And an exclusion list, which matters more than people expect. Without one you get autotune, trap hats and a modern pop mix bolted onto something that wanted none of them.

Four: the video, and the thing that breaks it

Video generation has one failure that swallows everything else: the same person does not stay the same person between shots. Generate twenty-eight scenes independently and you get twenty-eight cousins.

The fix is unglamorous. You write one locked description per character and paste it verbatim into every prompt, changing only the wardrobe. Mine run to about sixty words and specify jaw, hairline, build, skin texture and eyes. Skin texture matters more than it sounds: without an explicit note about visible pores you get a plastic sheen that reads as fake instantly.

Then each scene gets a shot, an action and a single camera move. One move. Ask for a push-in and a pan and a rack focus and you get something that lurches.

The maths of it

For a three-minute song with a video, honestly:

Reading the brief and finding the angle: 30 minutes
Lyric, draft plus rewrites: 90 minutes
Generating and choosing the recording, usually 20 to 40 attempts: 45 minutes
Character locks and scene prompts: 60 minutes
Generating and rejecting clips: 90 minutes
Edit, cut to the beat, colour: 60 minutes

Call it six hours of attention. Which is why anything advertising an instant personalised song for nine dollars is selling you the ninety seconds and skipping the six hours, and you can hear it.

The tools removed the cost of production. They did not remove the work of deciding what is good.