show index hide index
- After countless iterations of this process, the researchers observed a striking phenomenon: the artistic styles seemed to converge around twelve characteristic motifs. The images produced, while varied, displayed striking similarities. Scenes of lighthouses, elegant salons, and nighttime streets recurred tirelessly. These creations, often compared to
- An invitation to redefine originality
The recent study on the findings of AI-based image generators raises fascinating questions. While these technologies promise limitless creativity, their output is actually limited to just twelve artistic styles. By repeating cycles of creation, it becomes clear that AI favors familiar patterns and structures at the expense of originality. This observation forces us to reconsider the notion of innovation in a field where the machine sometimes seems more like recycling than invention. A thought-provoking study: The recent study on AI image generators has revealed a surprising reality: these tools, while impressive, are indeed only capable of producing twelve artistic styles. Paradoxically, these same generators promise infinite visual possibilities, but by their very nature, they tend to homogenize artistic creation. Through experimentation, researchers demonstrated that these models, instead of producing truly unique works, recycle and repeat the same motifs. A Limited Creative Mechanic The image models used to generate these creations rely on vast visual databases amassed on the web. This approach is based on combining existing references, which might seem like a true creative feat. However, the study showed that by overusing these references, the models ended up producing images with a narrow style, focusing on a limited number of visual expressions.The Visual Game of TelephoneTo explore this issue, the researchers implemented an experiment inspired by the concept of telephone.but visual. The creation model, named Stable Diffusion XL, received an initial description and produced a corresponding image. This image was then passed on to LLaVA to generate a new description, which fueled a new creation cycle.The results were revealing: from the very first iterations, the original image deteriorated as the cycle repeated. The transformations made the image’s appearance less and less identifiable, each iteration becoming further and further removed from its initial concept. The Convergence of Styles
After countless iterations of this process, the researchers observed a striking phenomenon: the artistic styles seemed to converge around twelve characteristic motifs. The images produced, while varied, displayed striking similarities. Scenes of lighthouses, elegant salons, and nighttime streets recurred tirelessly. These creations, often compared to
visual elevator music, represent soothing, but largely banal, compositions. Soothing but stereotypical creationsThis “visual elevator style” translates into unprovocative scenes that don’t assault the eye or challenge the senses. Indeed, these images echo a common taste, as if the AI were trying to please the widest possible audience. In this quest for visual accessibility, it’s not surprising that the process leading to genuine originality seems to be a source of confusion.
The difference with human creativity At this stage, it is crucial to highlight the striking difference between human creators and artificial intelligence. In a human process, each individual interprets the message in their own way, bringing enriching biases and nuances. AI, on the other hand, operates on fixed statistical preferences, acting as an echo of current aesthetic trends. It gives the illusion of creativity, but is limited to copying what is popularized by training data.
An invitation to redefine originality
This study paves the way for a broader reflection on what originality means in the age of AI image generators. While these tools can produce attractive visuals, their inability to truly innovate highlights the challenges related to understanding and teaching aesthetic sense. While machines excel at reusing popular visual elements, human creativity remains unrivaled and transcends stylistic boundaries. To learn more about AI image generators, consult resources such as this guideor this analysis on Stable Diffusion.
