I’m building a system for creativity. Read more about it here. Pre-order this week and save 50%. Offer expires Friday, June 7th. Get it now.

As generative AI’s various forms have evolved, they’ve mostly made a sudden leap from bad to good – or at least alright.

Text generation did it. Code generation did it. Image generation did it. And AI music just did it.

I checked out AI music late last year and then wrote this.

I’ve not ventured deep into AI-generated music. Why? Because. It’s. Awful.

Well, that was then. AI music made the leap to good.

The breakthrough is Suno, which is the GPT of music. You can hear a variety of samples at the site.

There are still many issues, but overall, its results are good enough, especially for instrumental tracks. Most importantly, it’s useful. 

I actually see a clearer path forward for AI music than AI images, which I still struggle to find a purpose for. AI music could be one of the most impactful forms of generative AI. Here’s why.

Music is very flexible, very plastic

Music is the most abstract of arts. It doesn’t necessarily need to sound entirely right in order to sound good to us.

Images are different. Images have to look right. If that hand has 6 fingers, ya blew it. The weird artifacts of AI music can become a style, just like record scratches and autotune and digital glitches did. 

Music can also be remixed with incredible flexibility. Working around its quirks is straightforward with the established software tools.

Generic music is useful

Generic is what AI is good at. Generative AI’s superpower is mediocrity. It can do fine, excellently.

Music is often a support player, not the star. For instance, most video content has music in the background. Unless you’re producing high-end projects, good enough music can work fine in your mix.

Stock music is gonna take a hit. If your business is based on making generic music, it gives me no pleasure to say that I think your business will need to fundamentally change.

A word of caution

I think something that many of us get wrong is this: we assume this velocity is normal.

We think leaps like the one AI music just made will happen again and again until AI’s are human level. But we’ve not seen these leaps recur yet. After that initial jump, development seems much more incremental. Don’t be surprised if we see the same here.

Will video be next?

The final media form that has not made the leap is video. This will be the toughest challenge. Creating a long sequence of images that all look right and also look right playing one after another at high speed – that’s a lot that can go wrong. And working around the glitches is not possible with current software.

What about Sora? The demos of OpenAI’s upcoming video generator looked great, but we’ve now seen actual work getting done with Sora and the results look a lot more like the wonky AI video we already have. Don’t be surprised if AI video’s leap doesn’t arrive with the public release of Sora.

Best!
Kirby

P.S. We also have deep discounts on exclusive coaching and mentorship opportunities. Book a session or sessions with me and I will help you do better creative work. These prices are ending this Friday as well.