I recently found myself watching a recording of Jean-Michel Jarre’s 2025 concert in Seville. Alongside the music and spectacle, there was a lot of discussion around AI and AI-generated content. That set me thinking, not in a grand theoretical way at first, but in a simple, curious way: what would ChatGPT visualise if I asked it to turn Oxygène into a single cartoon image?
Not “make a copy of the album cover”. Not “recreate Jean-Michel Jarre’s visual style”. Just: take the word, the title, the atmosphere it suggests, and see what happens.
That distinction mattered to me. Oxygène already has a famous visual identity, and the point of the experiment was not to imitate it or produce a substitute. It was more about asking what an AI system does when given a culturally loaded musical reference, a single evocative word, and a loose creative brief. Would it lean towards the known album imagery? Would it think of oxygen as science, atmosphere, breath, Earth, space, environment, or sound?
The first prompt was deliberately simple:
“Turn Jean-Michel Jarre’s Oxygène into a single cartoon image.”
Almost immediately, I realised that this was too open-ended. That was part of the interest. With image generation, the prompt is not just an instruction; it is also a negotiation. So I changed tack and asked the system to question me first rather than generate too quickly.
The questions were useful. Did I want it to feel like the original album cover or only loosely inspired by its atmosphere? Should it be abstract, surreal, narrative, or album-poster style? Should Jean-Michel Jarre appear as a cartoon character? Should the mood be psychedelic, playful, ecological, retro sci-fi? Should there be text? What format?
My answers became the real prompt. I said I did not want it to be like the album cover. I did not want Jean-Michel Jarre represented. I wanted an album-cover format, image-only, and I asked for a robot to be included. Everything else I left open.
That openness was important. I was less interested in controlling the result than in seeing what the system would choose to make visible. In other words, the experiment was not “can AI draw what I already have in my head?” but “what does AI think this could look like?”
Seeing the image changed the experiment. This is something I find interesting about working with generative AI: the output becomes part of the next prompt. Once there was a still image, I started to wonder whether it could move. Not a full music video, not an attempt to visualise the track in any literal sense, but a small animated object. Could the waves move? Could the image become a short looping GIF?
So the next stage became:
“Animate the waves and turn it into a 3 second GIF and save.”
That shifted the piece from imagined album cover to tiny animated artefact. The movement made it feel less like a finished answer and more like a sketch of an idea: oxygen as atmosphere, motion, flow, sound, and perhaps a little machine dreaming inside the scene.
Then came another turn. I asked for it to be changed “more in the style of Van Gogh”. Again, that raises questions. Referencing an artist’s style is one of the areas where AI-generated art becomes ethically and creatively complicated. In this case, I was not trying to pass anything off as Van Gogh, nor to copy a specific painting. I was using “Van Gogh” in the loose way people often do conversationally: swirling skies, expressive brushwork, heightened movement, intense texture.
But even that casual phrasing is worth pausing over. AI makes it very easy to ask for “in the style of…” and that ease can hide difficult questions about influence, imitation, authorship, and labour. This small experiment was never about copyright infringement, and it was never about replacing an existing artwork. It was about curiosity. Still, the process reminded me that curiosity does not remove responsibility. The prompts we use shape what is borrowed, echoed, flattened, or transformed.
What I enjoyed most was not any single output, but the back and forth. The first prompt opened a door. The questions narrowed the direction. The image suggested animation. The animation suggested style. Each step was prompted by seeing what came back.
That, for me, is where the experiment becomes interesting. AI image generation is often talked about as if it is simply a way to produce a picture. But in this case, it worked more like a visual conversation. I was not asking it to recreate Oxygène. I was asking what Oxygène might become when filtered through text, memory, association, and a machine’s visual habits.
I am still not sure what I think about the result. That may be the point. The image is not definitive. It is not “what Oxygène looks like”. It is one answer produced through a particular chain of prompts, choices, refusals, and accidents.
And, apparently, perhaps also a robot in a strange cartoon world, moving gently for three seconds.
Comments
Post a Comment