Domain Randomization (2025)
52 4K video clips of variable duration, randomly shuffled via media player.
Domain Randomization explores the production of synthetic image data sets via game engine environments and props. What does it mean to train image detection and generation tools for real world use on images produced in virtual worlds? Taking cues from Vilem Flusser’s Towards a Philosophy of Photography, the work explores the idea that black boxed technologies have consigned us to a world of endless reconfigurations of the same culture.
Domain Randomization exists somewhere between animated image data set, and video game world. Using thirteen pre-built environments sourced from asset marketplaces, a range of everyday objects and animals are randomly and rapidly rotated whilst lighting conditions change, producing hundreds of images that can be used to train AI image models. These image sets are then fed into object detection algorithms to check for accuracy, and the description of the scene is fed into a text to audio model to generate a synthetic soundtrack.