This Valentine's Day is the season of love, and what better way to celebrate than with a song? Now, thanks to AI tools like Suno, it's easier than ever to create music that expresses your passion.
Inspired by the story of a Spanish artist who married a hologram, I decided to give it a try to see how easy it would be to use AI to create not only a song but also a music video.
Using a combination of tools including ChatGPT, Suno, Leonardo, Stable Video, and MidJourney, we created a music video for the generated song “Echoes of a Digital Heart.”
Select a genre and generate lyrics
The first challenge was to come up with an idea for a song. I thought about writing something for my wife, who would have been married for 25 years this week, but in the end I decided on a fun song about unrequited love between a man and an AI.
The next step was to decide on a genre and have ChatGPT write the lyrics. I tried a few classic country love songs but they were awful and never mentioned again. Then I realized I needed to go for a more electronic feel and settled on 80s New Wave.
Unlike Beyoncé, who had huge success playing Texas Hold'em, AI doesn't have a country vibe to it, but she does seem to borrow the style of A-ha, Ultravox and other '80s icons.
After that, all you have to do is tell ChatGPT the basic concept of your story and the genre you want to cover, and ask it to generate lyrics for a song with two verses and a chorus.
“Echoes of a Digital Heart” includes the line, “In a dance of hearts and circuits and despair, I'm lost in your chords, your breathless air.” If this isn't love, then what is?
Making music with Suno

Suno is the best all-around AI music generator I've used to date. There are others on the market, and I think Google's MusicFX does the best instrumental work, but Suno adds audio to your songs.
I created a custom track in Suno, adding ChatGPT lyrics and my genre of choice. Since I can only create about a minute of music at a time, I split it into five segments to come up with a three minute song.
Each segment contains a verse and a chorus, or a pre-chorus and a chorus. After generating a clip, you can “Continue from this clip” to keep the music and style and only change the lyrics. Finally, you can “Get the whole song”.
Music video production

This was the most challenging part of the entire endeavor, and involved using a combination of MidJourney and Leonardo to generate nearly 80 individual images and then animating the best ones.
First, we had to create an outline of the story for the song. We asked ChatGPT to plan a story based on the lyrics and suggest an image (assuming a 5-second clip) that would fit that narrative.
It worked just as I wanted it to, even breaking down where to place the clips based on their timestamps within the song, so I took their suggestions, tweaked the wording and fed it into Leonardo. I used this Australian startup AI platform because of its photorealism and video generation.

Nearly 2,000 credits later, the video was about 90% complete, but two things were missing: a dream sequence and a singer.
For the dream sequences I used a new tool I will be reviewing soon: Stable Video from StabilityAI. Stable Video is a great video generator based on the Stable Video Diffusion model – the same model used for Leonardo Motion, but with more advanced control.
For the singer, we asked MidJourney to create a 1980s British style singer with a retro AI vibe, staring straight into the camera. They did not disappoint, although we did have to use MidJourney's zoom out feature to get exactly the result we wanted.
They then used Pika Labs to animate an image of the singer, moving their mouths to make it appear as if they were singing the song.
Putting it all together

The final step is to combine all the generated clips and music. Since I'm on a Mac and don't have the money to buy Final Cut Pro, I used iMovie to edit the last 3 minutes of the music video, which had all the tools I needed.
It's not the best song ever made, but the music video does have some moments of style changes between clips and the narrative gets a bit confusing, but overall I think it's good.
