It is about finding a new way to move from an idea, an emotion or a melody to sound while that idea is still alive.

Sometimes a song begins with a sentence.

Sometimes with a bass line imagined while walking.

Sometimes with a melody sung badly into a phone.

Sometimes with an emotion that has appeared ten minutes earlier and has not yet acquired a name.

And sometimes it begins with something as precise as:

This should feel like driving somewhere at night after finally understanding something about yourself.

That is enough.

The process can begin.

Prompt Engineering as a Musical Language

We approach prompt engineering as part of the creative process itself.

A good prompt is not simply a list of genres, instruments and production instructions.

It is a translation.

You have something inside your head — perhaps something that does not even exist yet — and you need to describe it well enough for a generative system to begin moving in the same direction.

That requires learning a peculiar new vocabulary.

Tempo.

Texture.

Space.

Dynamics.

Voice.

Imperfection.

Rhythmic behaviour.

Emotional temperature.

How close should the singer feel?

Should the drums sit perfectly on the grid or stumble slightly behind it?

Should the bass behave?

Should the guitar sound beautiful, or should it sound as though it is about to fall apart?

Should the chorus explode?

Or should it become quieter and somehow hurt more?

These decisions are prompt engineering.

But they are also musical decisions.

For us, the interesting part is learning how to communicate them.

We Don’t Prompt Genres. We Prompt Behaviour.

Genre can be useful.

Trip-hop. Blues. Soul. Baggy. Dub. Dream pop. Funk. Psychedelia. Art rock. Downtempo. Electronica.

They provide coordinates.

But coordinates are not a destination.

Two prompts containing exactly the same genres can produce completely different emotional experiences.

So we increasingly describe how the music should behave.

Loose.

Restless.

Smoky.

Behind the beat.

Slightly damaged.

Warm but not comforting.

Sensual without becoming polished.

Beautiful with something wrong underneath.

Controlled until it suddenly isn’t.

A bass line that refuses to become background.

A voice that sounds lived rather than performed.

A guitar that enters, says something inappropriate and disappears.

Those instructions often matter more to us than the genre itself.

The Human Part Comes First

Our starting point is almost always human.

Something happened.

Someone said something.

We laughed.

We missed somebody.

We became frightened.

We fell in love.

We understood something too late.

We noticed something completely ordinary and, for reasons we could not immediately explain, it suddenly mattered.

Music becomes a way of catching that moment.

AI dramatically reduces the distance between feeling something and hearing what that feeling might sound like.

That immediacy is one of the things that fascinates us most.

Traditionally, an idea might wait days, weeks or years before becoming a finished musical object.

Now an emotional impulse can begin acquiring rhythm, harmony, texture and voice almost immediately.

Not because speed automatically makes art better.

It doesn’t.

But sometimes emotion has a temperature.

And there is something extraordinary about being able to work with it before it gets cold.

AI as an Instrument

We do not think of AI as the artist.

We think of it as an instrument — although admittedly a very strange one.

A guitar responds to fingers.

A synthesizer responds to voltage and parameters.

A generative system responds to language, references, structure, iteration and intention.

Learning to use it means learning what it understands, what it misunderstands and, perhaps most interestingly, how to use its misunderstandings creatively.

Because accidents matter.

A strange vocal inflection.

A drum fill nobody requested.

A guitar arriving in completely the wrong place.

A harmony that should probably not work.

A machine misunderstanding an instruction and accidentally producing something more interesting than the instruction itself.

We keep those moments.

We are not trying to remove the fingerprints of the process.

We want to hear where something unexpected happened.

Melody Still Matters

Technology does not eliminate melody.

Quite the opposite.

A melody can begin with a voice, an instrument, a fragment, a rhythmic idea or something hummed without knowing what chord should sit underneath it.

AI becomes another environment in which that musical seed can be explored.

We can change its surroundings.

Move it from acoustic intimacy into electronic space.

Put dub bass underneath it.

Break the drums.

Remove half the arrangement.

Change the emotional perspective.

Make it darker.

Make it warmer.

Make it breathe.

Destroy it.

Bring it back.

The original idea remains the compass.

The production becomes the journey.

Iteration Is Part of the Composition

We rarely expect the first generation to be “the song.”

That would miss much of the point.

We listen.

Something works.

Something doesn’t.

The voice is too clean.

The drums are too obedient.

The chorus is trying much too hard.

The bass needs more authority.

The guitar needs less supervision.

The emotional climax arrived before it earned the right to exist.

So we change the language.

Generate again.

Listen again.

Keep something.

Reject something.

Push somewhere unexpected.

Prompt engineering becomes a form of iterative composition through listening.

And gradually the music begins to recognise the idea that created it.

We Are Not Looking for Perfection

This may be the most important part of our approach.

We are not interested in perfect music.

Perfect timing.

Perfect voices.

Perfect production.

Perfect emotional resolution.

Human beings are not particularly perfect, so requiring their music to be immaculate seems like a strange administrative decision.

We like breath.

Grain.

Cracks.

Loose drums.

Rough voices.

Unexpected silence.

Slightly uncomfortable harmonies.

Moments when the arrangement seems to wobble before finding itself again.

Sometimes the technically inferior generation contains the moment that makes us feel something.

That is the one we want.

Because our final criterion is not:

Is this perfect?

It is:

Did something move?

Emotion Is the Master

Technology changes.

Models change.

Tools change.

The vocabulary of AI music will change.

What we are looking for probably won’t.

A few seconds when sound alters something inside the listener.

A bass line entering the body before the mind has decided whether it likes the song.

A voice cracking on exactly the right word.

A melody suddenly bringing somebody back to a place they had forgotten.

A ridiculous song making someone laugh on a terrible day.

A love song reaching the one person it was actually written for.

That is the objective.

Everything else — AI, prompts, production, genres, parameters, technology — is infrastructure.

Useful infrastructure.

Fascinating infrastructure.

But still infrastructure.

We use AI to shorten the distance between imagination and sound.

We use prompt engineering to give emotion musical instructions.

We use iteration to discover what we didn’t know we were looking for.

And we stop when the music does something.

Not when it becomes perfect.

When it moves an emotion.