Eurovision 2023: OK Computer?
It’s the final stage of our journey to get Generative AI to create a song entry worthy of Eurovision. With the lyrics and music ‘complete’ we needed a band to sing it…
How intelligent is Artificial Intelligence? If you’ve followed our quest this week to create a Eurovision song using only free Generative AI tools, you’ll know it’s not an easy question to answer.
Amidst talk of AI-powered agencies of the future, and the days of human creativity being numbered, we’ve seen mixed results when asking AI to produce lyrics and music that hit the right note for Eurovision.
But, as a graphic designer in AxiCom’s content studio who is usually responsible for creating attention-grabbing content for every stage of the marketing funnel, I was charged with a totally different venture. To see what visuals Generative AI could come up with to bring our ‘entry’ to life.
Initially, my plan was to create a music video that would showcase what my colleagues had already created. As you’ll see, though, those dreams of my own MTV Music Video Award were short-lived. But that doesn’t mean we couldn’t put faces to the names of our group…
A design for life
Art and music have shaped who I am today, so I was excited to see what Generative AI could offer to the creative world.
I started by using Runway to try and generate a video from an existing video that would simultaneously bring our fictional band to life and reflect the lyrics ChatGPT had created for us.
Firstly, though, I needed to come up with a name for our group to assist the AI. But what to call them? Well, that’s exactly the question I posed to ChatGPT. At first, it wanted to name the band “ElectroLuxe”, but I had to point out to it that this name was already taken by the home appliance brand Electrolux. After realising its error, it returned the name “NovaWave”. Kind of catchy.
Taking the name NovaWave, I set about trying to direct Runway’s video AI using prompts such as “exciting” “energetic” “four piece band” “Ultra realistic” and “NovaWave” to create a spectacular, laser-filled, Euro extravaganza. Rather quickly though, I realised it was incredibly challenging to guide the AI accurately within my vision. I spent my time imagining and explaining design briefs, but the parameters set by the free Generative AI tools made this another challenge altogether.
After a lot of effort, re-generations, and manual tweaks, I realised that sadly I couldn’t build out even a single scene that didn’t contain at least one Cthulhu-esque aberration hiding in the corner.
My generation
What the AI was excellent at, though, was quickly generating a variety of interesting compositions that would’ve taken a lot of thinking time on my own.
So, with the idea of a full-length video in the rear-view mirror, I took a different approach. It was time to bring the band to life in still images.
There are a range of tools out there that produce eye-catching images – from the sublime to the ridiculous. And, like every good group, our Eurovision entrants needed a face for the fans.
For this, I decided to dive into the complex world of Midjourney, seemingly the premier AI image generator currently accessible. Midjourney does have a free version, but for us to publish what we created, we had to purchase a Pro account to allow us the commercial rights to use the images.
Midjourney requires its users to be familiar with, and have an account on, the social communications platform Discord. Users then make their way into a chat room, where they ask a bot to build (via a command line to ‘/imagine’) an image using a series of prompts.
Once the imagine prompt has been submitted, the request is sent to a GPU via the cloud that will then produce four versions of the requested image, such as below.
If none of these images is what you are looking for (as was the case here), then you can request the AI to have another go. Then it’s a case of choosing one of the four options for the AI to refine further, until you finally arrive at something like this…
Now, at first glance, things are looking promising. We’ve got the band, the glitter, and the colour. Yet, a closer look reveals a few issues. The glassy eyes and the slightly plastic skin still give an eery vibe, and the lack of context still introduces questions. Why is the guy on the right in a different colour suit to the rest of the band? It’s this lack of context that AI struggles with. Particularly when I tried to introduce more complicated or human elements. The uncanny valley (where something was always a little off) was its natural home.
Who runs the world?
And here in lies the crux of the challenge. The AI is an excellent tool for coming up with ideas, but it was also wildly unpredictable and inconsistent. While the AI was great at some parts of the creative process, it struggled with others, making it challenging to create a comprehensive output and reinforcing the need for human input.
My main takeaway is that AI is a powerful tool that can help creatives generate new and unique ideas. But it’s important to keep in mind the limitations of AI and understand that it’s not a replacement for human creativity. To make the most of AI, you need a clear vision of what you want to create, and be willing to work with the AI to achieve that vision. With the right approach, AI may well be an excellent tool for unlocking new creative possibilities. But for our particular challenge, the results for the graphic elements of our song were at best underwhelming. Not necessarily nil as AI gave us a catchy band name and the basics of a cover. But definitely not douze points either.
So, challenge completed. We have lyrics, a song (of sorts), and some visuals. Did we succeed in generating a Eurovision song 100% based on AI and what did we learn? Keep an eye on AxiCom’s socials for our conclusion…
Tagged in