For my submission, I’m sharing Krungthep 2.0, a short AI-assisted animation that’s part of an ongoing series exploring how machine vision imagines the future of cities. This piece focuses on Bangkok, my home city, and uses AI as both a tool and a lens to reconstruct speculative urban landscapes based on digital memory. I wanted to explore what happens when a city is reinterpreted by algorithms trained on public data, media saturation, and fragmented visual culture. What does the future look like when it’s not shaped by people, but predicted by machines? In this project, I used Midjourney to generate imagined cityscapes of Bangkok. From there, I made a personal creative rule to rely only on AI-driven processes to build the visuals. I took those static images and used RunwayML’s camera control feature to give them depth and movement. RunwayML wasn’t the only AI tool I used, but it played a pivotal role in speeding up the animation process. I could have spent hours modeling scenes in Blender or animating frame by frame, but I chose a method that felt more efficient and aligned with the concept. The goal was not realism, but immersion, and I wanted to create the feeling of stepping into an AI’s imagined world. Once I had those motion-augmented clips, I used Luma3D to generate Gaussian Splat reconstructions. I chose Luma3D because of what I learned from an earlier project, Latent Space. During that experiment, I noticed that Luma3D often produced fragmented, unstable environments that felt like half-formed memories. Instead of avoiding that limitation, I leaned into it. The glitchy geometry and incomplete renderings felt symbolic of how AI perceives space. It became a visual metaphor for how speculative and disconnected AI-driven futures can be. After generating the splats, I manipulated them directly by editing the motion, opacity, and color of the individual points through custom code. Although I realized later that I could have used Unreal Engine, which supports Gaussian Splatting more smoothly, I chose Unity instead because I wanted to challenge myself and better understand how splat files are structured and rendered. I designed a dark, empty scene with deep red tones to represent a digital abyss. For me, that space captures how I imagine AI vision: unanchored, speculative, and shaped by fragmented data instead of lived experience. For the sound, I composed the audio using a mix of sound presets and procedural sound generators. I wanted the sound to echo the eerie, synthetic tone of the visuals and create a sense of quiet tension throughout the environment. The result is an ambient experience that is more about atmosphere than narrative, and more about suggestion than resolution.