Isopolis is a reflection on automation, perpetual growth and the ways in which artificial intelligence is contributing to the commoditization of creativity. In Isopolis, an infinite cityscape is generated from a random combination of buildings pulled from a large dataset of AI-generated images. The website drifts across this cityscape, getting its direction from mouse input. The viewport slowly moves closer or further away from the viewer over time via a random process.
The underlying image dataset is created using an automated process put in place by the artist. First, an image is generated from a prompt into a text-to-image AI model. The resulting image is passed on to another program which leverages a convolutional neural network trained for object segmentation and background removal. The image is then fed to a third program which automatically crops and algorithmically determines the angle of the bottom corner of the generated building. This final program also performs a selection, ensuring that the image has a suitable composition within the frame and that the isometric projection angles are within acceptable tolerance limits. Unsuitable images are discarded.
The result is a theoretically infinitely scalable dataset of images, limited only by available storage and computing power. By adopting an industrial approach to asset generation, I aim to explore how computation, specifically AI tools, can facilitate the creation of visual material at a scale unattainable through manual labour alone. The sheer number of generated images, combined with the randomization of their composition, makes it impossible to predict the final appearance of the city until the software is executed. Simultaneously, the work questions the value of applying automation and industrial processes to the creative process. What are the costs? Is quantity a worthwhile tradeoff for quality? Isopolis, with itsinfinite urban sprawl, quickly devolves into a dystopian landscape where the uniquely generated buildings occasionally reveal quirks and glitches in the data. More importantly, the promise of “infinite content” gradually gives way to a monotonous homogeneity.