What lies around the corner

I’m revisiting a sequence of three posts from February-March 2014: #184 Turning the corner, #185 Humanities in the wild, and #186 Betwixt and between. They each fit a theme that later informed the writing of my book Network Nature: The Place of Nature in the Digital Age, published in 2018.

Whether in nature or architectural settings, threshold events constitute significant moments — crossing a river, reaching a summit, passing through a doorway, turning a corner. On my re-reading I don’t think I can expand on the case I made in those posts of 12 years ago.

I mentioned various technical mediations such as cinematography, video gaming, VR (virtual reality) and EEG (electroencephalography) as means to access, represent, amplify or problematise the psycho-spatial experience of traversing boundaries.

In the first of these three posts I included a photograph of the entrance to Dartmoor Prison in Devon, England. There’s a heavy stone archway leading to an open vestibule and an imposing door beyond. The grim spatial configuration invited curiosity, not only about what lies beyond the door, but what we see if we pass beneath the arch and turn the corner (left or right) to see what lies in the courtyard.

In the intervening years we can add the technology of AI-generated video that is able to create an impression of what might lie around the corner, at least as space is revealed photographically.

To demonstrate how AI can simulate such an experience I have selected a photograph more appealing than a prison entrance. Here is a presentation of one of the many portals at the partially restored temple of Banteay Srei in Cambodia that I visited in 2024. It’s part of the Angkor Wat complex. The temple presents spatially as a succession of portals aligned across a series of enclosures.

Interestingly, the visitor would catch sight of other visitors as they criss-crossed this intricately carved red stone labyrinth. I managed to capture one such framing in a photograph.

In light of the seductions of AI-generated video, I prompted RunwayML to “Animate this scene.,” adding “The camera moves through the opening as the people move out of the away. Importantly, the camera turns left to reveal a new setting.”

It took five attempts to achieve what I was after. The AI seemed to have difficulty passing through the portal. The presence of people may have confounded my prompting. In any case, the rendering of the people and their movement strikes me as even more compelling than the presentation of the architecture.

In RunwayML’s text-to-video workflow, the AI elaborates on my prompt, which it translated as, “The camera moves straight through the ancient stone opening, passing the people who gracefully move aside. Once entirely clear of the opening, the camera executes a smooth left turn, revealing a vibrant, previously unseen setting with lush foliage and distant ruins under a clear sky.”

So the process begins with a text-to-text translation that gives the video simulator something to work with, introducing colour, camera controls and other visual cues. (For a later investigation I’ll look at how the AI diffusion model might “hallucinate” without such intermediate prompting.)

Here is a video that captures that description. I added in the failed attempts as well. Each sequence begins with my original still photograph before it springs into life. I also follow the sequence with a more surreal prompt, “The camera moves straight through the ancient stone opening, passing the people who quickly move aside. Once entirely clear of the opening, the camera executes a rapid left turn, revealing a surreal futuristic urban setting.” With its focus on the people, note that my prompt included nothing about panic.

Bibliography

• Coyne, Richard. Cornucopia Limited: Design and Dissent on the Internet. Cambridge, Massachusetts: MIT Press, 2005. 284 pages.
• Coyne, Richard. Network Nature: The Place of Nature in the Digital Age. London: Bloomsbury Academic, 2018.