Bringing a murder ballad to life with AI
My tool box
In 2022, I wrote the lyrics to a femme-fatale dark country song that started off like this: “Trash bag full of all my clothes, guess it’s time for me to go.” The year prior, I learned how to play the ukulele and, after learning 4 chords, thought I was hot shit enough to write my own music. However, writing a murder ballad with a ukulele didn’t hit the way I thought it would.
from rough notes to full production
Fast forward to the last 2 days: My husband has been OBSESSED with this Suno Gen AI music app and reviving every song he’s ever written. Inspired by his enthusiasm and now able to play a guitar, I thought why not, let’s make my own snotty infant of an idea into a real grownup song.
I started by revisiting my lyrics and song structure before touching AI at all. The first verse came from giving myself permission to explore:
“I’m a black belt at picking fights
A rolling storm when you’re in my sights
From blue skies to tears and lies
Take a hit, love, you’ll be alright”
Then I asked Chat GPT analyze the song and make suggestions, explaining what I feel the core message of the song to be. I took what felt good (ex: “neon lights”) and ignored the rest.
From there, I experimented with new chord progressions, and landed on something beginner friendly: Am E G D repeated and with a simple down-down-up-up-down strumming pattern.
But the bridge needed tension and distinction. I started a new chat and asked GPT again for a few options. A new chat ensured the model was focused on what mattered most - the musicality of the progression - rather than the lyrics or other noise from the previous chat. It spat out 4 options with explanations of why each might work, and I tested all 4 on my guitar before deciding. Option 2’s progression of Dm – F – C – G sounded the best with the song.
Putting it all together, here’s what a really really rough demo of me singing and playing the original sounded like:
I sent my dirty demo to my husband and let his obsession take over. We created about 10 different versions of the song on Suno, adjusting the prompt, weirdness level (which is similiar to ChatGPT’s temp variable), and pace until it felt right. Here’s an early version that had a weird warble on the intro and lacked the “umph” we were looking for.
A few more adjustments, and the song you see at the top is the result! AI and humans, going 0-1 to create high fidelity demos in an evening. I found this workflow to be helpful at speeding up the raw idea-to-shareable demo process, as well as being a suitable substitute when you lack a collaborator to bounce ideas off of.
Most importantly, my partner and I had a blast playing, learning, and experimenting together. Sharing creative energy and playfulness is rare while we’re long-distance, so finding this new way to connect is a fun surprise. Now if I can only convince him NOT to make his “Nigel the gecko needs a woman” song.
<3EQ