One of the great selling points of AI is speed.
Apparently I missed that memo.
I could probably have released hundreds of songs by now.
Instead, I have spent hours fixing tambourines.
I’ve regenerated songs because one line didn’t land properly. Changed a lyric after hearing it sung. Tried another model. Tried another voice. Pulled songs into stems. Put them back together. Changed the arrangement. Changed it back. Learned bits of a DAW because I couldn’t get something to sound the way I could hear it in my head.
Very efficient.
AI can make a song incredibly quickly
That's true.
Give an AI music generator a prompt and some lyrics and you can have a complete song in minutes.
And sometimes what comes back is astonishing.
That's one of the things I love about this technology. Someone like me, who doesn't play an instrument and can't sing, can hear an idea that previously existed only in my head become music.
That's extraordinary.
But there's a slight problem.
Being able to make a song quickly doesn't mean I've made the song I wanted.
Those are two very different things.
The first generation is usually where I start
Sometimes I get lucky.
A voice appears that I love. A melody catches me completely by surprise. AI does something I would never have thought to ask for and the song suddenly becomes better because of it.
That's the collaboration bit I find exciting.
Other times it's nearly right.
And we've already established that nearly is dangerous.
I'll change the prompt.
Change the lyrics.
Try another version.
Take the best part from one and combine it with something from another.
Decide version four had more feeling than version twelve.
Wonder why I didn't just leave version four alone.
Go back to version four.
Discover version thirteen.
This is my highly sophisticated production system.
More isn't necessarily better
There's an obvious temptation with generative music.
If you can make ten songs, why not make a hundred?
If you can release a hundred, why not release a thousand?
And people do.
I understand why. Streaming platforms reward catalogue, AI removes many of the old barriers to making music, and suddenly scale is possible in a way it never was before.
But that's not particularly interesting to me.
I don't want the achievement to be how much music I managed to generate.
I want somebody to actually care about one of the songs.
I'd rather someone play a song twice than scroll past twenty of mine once.
This is where the human bit matters
People often talk about human involvement in AI music as though it can be measured by counting how many things a person physically did.
I don't think that's the interesting part.
For me, the human part is choice.
Why this lyric?
Why that voice?
Why does this version make me feel something when the other one doesn't?
Why does the chorus need to come back differently?
Why does that sound annoy me?
Why does the tambourine need to drive rather than sit politely in the background?
Why have I listened to version 27 twelve times when version 14 was clearly better?
AI can give me possibilities very quickly.
It can't decide which one matters to me.
Taste might become more important, not less
The more AI can generate, the less impressive generation itself becomes.
That's inevitable.
If everyone can make something, then making something isn't the differentiator anymore.
Choosing becomes important.
Editing becomes important.
Having something to say becomes important.
Knowing when something is finished becomes extremely important.
I haven't mastered that last one.
Obviously.
But I think that's where this gets interesting.
The technology may make music abundant.
It doesn't make taste abundant.
I don't want to make content
This is probably the biggest thing I've realised.
I don't want to make music simply because a platform needs another upload.
I don't want to generate fifty tracks so I can feed an algorithm.
I want to make records.
Even if they're made in a completely new way.
Songs with characters behind them. Lyrics somebody remembers. Production details nobody consciously notices but that change how the whole thing feels.
Songs that might sit unnoticed for months and then suddenly find the person who needed one.
That takes longer.
I'm okay with that.
So yes, AI saves me enormous amounts of time
Which I then use to create entirely new ways of wasting it.
I can generate music in minutes and spend days deciding whether I like it.
I can create a singer who doesn't exist and then worry whether her latest song actually sounds like her.
I can build a 1967 girl group from nothing and subsequently lose an evening to a tambourine.
This may not be the productivity revolution everyone had in mind.
But I'm making music I couldn't have made before.
And people are listening to it.
For now, that's a much more interesting measure than how many songs I can upload.
AI made making music faster.
It didn't make me want to rush it.
