American Pie as told by 'Banksy'
Can we use artificial intelligence to interpret lyrics and tell a story?
On our morning rides to school, my 7-year-old daughter has recently started playing Don MacLean’s iconic 1971 ballad, American Pie. It’s a long song. Like a really long song – which is perfect for our 10-minute drive.
She likes it because we sing along. I like it because it takes me back to college parties when everyone would belt out the lyrics, which mourned a generation’s loss of innocence. Not ours, mind you, but we knew our turn was coming. I remember soaking up the sentimentality and dramatically thinking, “This song is going to bring me back to this moment for the rest of my life.” And it does.
Trying to explain the lyrics to to a 7-year-old is another matter, entirely. I told her the song was about Buddy Holly and how he died in an airplane crash in 1959. (I immediately realized that while she has flown many times, I had just introduced her to the fact that planes can crash. I quickly added: “It was a small plane, and flying was not as safe back then as it is today.”)
It got me thinking about MacLean’s lyrics. I knew some of the obvious references: “The day the music died” was Feb. 3, 1959, the day Holly’s plane crashed in Iowa, killing everyone on board, including Ritchie Valens and J.P. “The Big Bopper” Richardson. I assumed that “While Lenin read a book of Marx, the Quartet practiced in the park,” was a reference to John Lennon and the Beatles.
I googled the song and read some of the different interpretations offered over the years. (It sure seems MacLean had a strong dislike for Bob Dylan.)
The other thing I learned was that MacLean, who is still alive, set some of the record straight in 2015 when his original working manuscript sold at a Christie’s auction for $1.25 million. His explanation of the lyrics are in the auction catalog notes. If you want to dig into that, there’s an entire Wikipedia entry on it.
But what if Midjourney was asked to interpret the lyrics – 8 minutes and 42 seconds worth of allusions and obscure references? To make it more interesting, I asked the AI to do it “in the style of Banksy,” who for all we know is actually a cyborg ‘living’ in the UK.
In many cases, the Midjourney AI – which doesn’t have the context of MacLean’s explanations – was extremely literal. For the stanza, “Helter skelter in the summer swelter; The birds flew off with a fallout shelter; Eight miles high and falling fast; It landed foul on the grass, the players tried for a forward pass; With the jester on the sidelines in a cast.” Midjourney returned an image that tried to incorporate everything it was fed.
In other cases, it produced remarkably poignant images that captured the mood of the prompt.
One of the things I noticed was that when the prompt contained a lot of fragmented thoughts, which was often, the AI’s storytelling was weakest. I imagined the AI throwing up its arms and saying, “WTF” am I supposed to do with this? And the longer the prompt, the more likely for the generated image to stray from the Banksy style.
I reworked some of the images by breaking up the stanzas into two or three lines per prompt. That seemed to work with a few, but for others, the generated images were no better (or worse).
AI can’t replace human designers (yet)
As a visual storyteller, my biggest takeaway is that we have reached a point where AI-art platforms, in some instances, can replace human graphic artists or photographers. If you’re creating photo illustrations for blogs, social media, email campaigns and other digital products, an AI-art generator may be all you need.
If you need exact details illustrated a certain way, or you’re insistent that the people in your illustrations have five fingers, instead of six or more on each hand or that text make sense, then you still need a graphic designer to oversee the process.
As someone who has worked closely with graphic artists, the process for creating graphic designs is very similar whether you’re dealing with AI or a human: You explain your concept the best you can and they / it comes back with an initial design.
But here’s where humans still hold the big advantage (for now). When it comes to making revisions, refinements and changes – it’s no contest. Give me a human designer any day. If an illustration is critical or important, I need a human involved in the process.
As for Midjourney’s interpretation of American Pie?
I will say that in almost every case, the four images in each initial gird were all very good. For this exercise, I tried to pick the image that best fit the mood of the lyrics and best incorporated Banksy’s style. So, even though the AI was generating the work, it was a human-machine collaborative process, with a writing credit to Don MacLean and artistic inspiration from Banksy.
Below are some of the images that the AI generated (If you want to see all 25 images of the song, you can find them on my website).
All images were generated by Joe Newman using the Midjourney platform. You can also follow along on Instagram at @most_sublime_media. Get fresh content delivered to your inbox by subscribing to our Substack newsletter.