How AI and I create the songs

Get started with AI music generation

If you love music but don’t have the time, talent, skill, or whatever to produce your own music, give an AI music generator a try. I use Suno AI. I have the paid Pro plan, which is the mid tier of Suno’s plans. They have a limited free plan, Pro, and Premier, which is the choice of professional musicians. The free plan is good for dabbling and seeing how it works, and whether it’ll be worth it to you to pay for a higher tier. I've easily had enough fun so far to make it worth it.

Most actions in Suno require “credits.” For example, generating two examples of a song will use up 10 credits. I think the free version starts you with 50 credits per day. The Pro plan currently gives 2500 credits that recycle each month, with no carryover of unused credits. There are options to purchase additional credits.

You can input much of the music yourself if you have the capabilities. Or, like me, you can write the lyrics and decide how you want the AI to set it to music through prompts, settings, and edits. Or, if you don’t want to put much creative effort into it, you can have the AI generate both the lyrics and music, or just instrumental music. 

Here's a good summary by musician and producer Rick Beato (7:18 minutes long) of where AI music stands today. Spend the day reading the comments to that video and you'll have an idea of the controversy surrounding AI music. Like it or not, it is the future. You may as well bend it to your own will.

How to generate music

Above: Suno AI's user friendly interface

Let’s just jump in and see how I create a song.

First I write the lyrics with a particular style of music, beat, tempo, feel, instrumentation, and structure in mind. I also include any lyrics that I want backup singers to sing after a line (putting those in parentheses). For example, from The Watering Hole:

Old Mr. Jackal came a dragging on by (Shooby dooby dop bop, Shooby dooby dop bop)

Tail hanging limp cause his throat was so dry (Shimmy dimmy wop pop, Shimmy dimmy wop pop)

Next, I go into Suno AI and input the lyrics.

Then I enter a short prompt describing my vision of how I’d like the AI to produce the actual music. This can include stylistic comments, tempo (beats per minute), and descriptions of the voice I want and instrumentation. The app will also give you suggestions if you want them. I keep mine short because I want to give the AI the freedom to improvise. Here’s an example of the prompt I used for Hopping Away on the Stars:

Magical pop folk, storytelling, lightly rolling and happy rhythm, flute and acoustic band, down home vocal style

I then adjust the Advanced parameters, which include choosing a male or female vocalist (or none), and two sliders that are labeled “Weirdness” and “Style Influence.” Higher Weirdness settings allow more improvisation and creativity on the part of the AI. Higher Style Influence settings keep the style more in line with what is in your prompt. In practice, I have found the results of both sliders are inconsistent. So far I find settings around Weirdness 60 and Style Influence 50 or 60 produce the most pleasing results. 

Then I name the song and hit the “Generate” button. The app generates two examples for each button press. I’ve found that usually the first few generations are my favorites, although if I haven’t been clear enough in my concept of the music initially or didn’t write my prompts well, I may have do up to 20 or so generations before I get something I like. 

Once I have a version that I like, and often that includes some “happy surprises” that diverged from my original concept but make the music better, it’s time to do any edits. Edits can include replacing sections, replacing particular lyrics, “covering” it with a little different feel and instrumentation, and various other tweaks. Some of the edit tools pretty much don’t work and others are problematic if they do work. While frustrating, that’s the way AI development is. It’s in its early stages, and we have to expect some poor functionality, hallucinations, or other aberrations. 

If the in-app edit tools are not working, I’ll sometimes put the song into a tool like Audacity and do some editing in there.

Instead of using the generated thumbnails for posting on my blog, I’ll use a free image generator such as that at Perchance.com. These free versions can have some really weird output (three-armed people, etc.) but that’s part of the fun. I’ll sometimes make some edits to the AI image in a graphics app.

What does it take to generate music you’ll like?

The more you know about music, the more likely you are to come out with something that you want to listen to. I was a music major in college and a part-time working musician when I was younger, so I have some knowledge and experience to help me. But if you have a sense of what you like and can analyze what it is about it that you like, you are on your way toward coming up with satisfying results.

It’s a good idea to know how most popular music is structured, with sections called Verse, Chorus, Bridge, etc. Here’s a good basic explanation of these structural elements. Whether you write your own lyrics or not, and I strongly urge you to give that a try, prompting the AI with structure will get the song closer to what you envisioned.

Knowing the names of genres and sub-genres that you like will also help guide the AI. There are many resources online that cover this. When writing a prompt, it helps to think not only about genre or style, but also the way you want it to sound: gritty, triumphant, light-hearted, etc. And think of what instruments you want in the song. Depending on your input settings, the AI may ignore some of your input, but often it improves the output. I love to hear what it comes up with. Think of it as a collaboration between you and the AI. Weird, huh, but that's what it is.

What to do with your music

I generate music for my own pleasure, both helping to create it and listening to it. I’ll share it with family and close friends, and occasionally with a specialty group if the song is aimed at them and I honestly think they might like it. Don’t be surprised, however, if you encounter some people who are totally against AI music and think it’s all just “slop.” It’s not, but that won’t be the general consensus until we get used to AI in all the aspects of our lives.

Who owns the generated songs? Well, whatever part you wrote is yours to  use, but it’s not quite that simple. AI is based on learning from lots of copyrighted music, so the legal issues are yet to be ironed out. The tool you use will have conditions of use that explain what you can do with your music. Aside from that, if you publish your music in any way, there’s a chance if it’s good enough that someone may claim it’s theirs. So the best way to approach it, in my opinion, is just to accept that it’s not fully your music and enjoy the creative process and listening to it.

I discourage people from trying to monetize their AI music. Why? Because most casual users aren’t going to make anything substantial off of it, and frankly, monetizing everything eventually destroys it. Just enjoy it for yourself. There are probably millions of AI generated songs out there already, so no one besides your friends and family probably wants to hear yours. My blog is mainly for me, my siblings, and a few close friends. I’m only writing this piece in case any of them or someone who stumbles on my blog is curious. Plus, I enjoy writing.

Ethics

The whole legal and ethical framework underlying AI generated music is up in the air. My only recommendation is to be honest about your part in the creation of the music. Don’t misrepresent the AI contribution as your own. For example, while I write my own lyrics and try to guide and prompt the AI to deliver something similar to what’s in my head, in the end the music part is AI generated. I may casually call it “my song,” but the reality is it’s an AI song using my lyrics and some input from me.

Remember that any music that’s even in part generated using AI is really the product of countless musicians whose music has been fed into the AI, in most cases without compensation or even permission. That’s something which, as a former musician, I hope changes in the near future. I think it is already changing. It also bothers me to see AI generated music passed off as human music and monetized on platforms such as YouTube, Spotify, and Apple Music. I’m fine if it shows up in my feed, but it should be clear if the music is partly or wholly generated by AI.

I'm having a blast with this, and encourage you to try it yourself!