Visions Of Destruction

2023 Interactive installation (Stable Diffusion, eyetracking) Technical realisation: Mar Canet and Isaac Clarke

The interactive artwork "Visions of Destruction" presents an immersive real-time AI-driven experience that portrays humankind's devastating impact on terrestrial ecosystems. The viewers' gazes are seamlessly detected through an eye-tracking system, in the viewed areas unfold captivating exploration animations of the consequences of our actions on the environment. The simple act of observation results in transformations of the AI synthetic landscapes generated using Stable Diffusion. With each fixed gaze upon the landscapes, the viewers inadvertently contribute to their visual pollution and destruction. However, amidst this sombre reality, a glimmer of hope emerges when users graciously grant the landscapes moments of respite by averting their gaze. It is in these fleeting instances, as indicated by the elapsed time when eye-tracking fails to detect any viewer, that the landscape begins its remarkable journey of regeneration. Through this symbiotic dance between human presence and absence, the landscapes find solace, allowing their inherent beauty to flourish once again. Let us recognize the profound impact of our actions, embracing the responsibility to protect and nurture these precious natural wonders.

Living in an era marked by disorder, climate crisis, inequality, and the threat of systemic collapse, we are driven by an urgent need to address these existential challenges. Our artwork mirrors this urgency and disorder, harnessing the vast power of AI to illustrate the potential devastation of our ecosystems. "Visions of Destruction" encapsulates deep time in its evocation of nature's splendor, while acknowledging the harsh reality of the irreversible damage we've inflicted on our environment. This piece brings to light the tension between the idyllic beauty of nature and the dystopian landscapes created by human activity. As viewers interact with the piece, their gaze initiates a transformation in real time. Through this intimate interplay, we seek to inspire an understanding of the gravity of our actions and the fleetingness of time - finite, invaluable and irretrievable. Each gaze leaves an indelible mark, echoing the profound effect we have on our environment. The installation software is built using Python and incorporates various open-source libraries along with custom-developed software. The AI Python code is built upon PyTorch libraries, primarily utilizing Stable Diffusion with Diffusers and Frame Interpolation for Large Motion models. Additionally, we leverage OpenCV to generate masks for eye-viewing areas. One noteworthy feature of our eye-tracking system is its automatic calibration-free functionality, which seamlessly detects the person most engaged with the sensor.