a Python-powered “prompt-morph” demo that harnesses a latent diffusion model to weave AI’s visual interpretations into one continuous, dream-like journey.
By feeding in the prompt “Song Dynasty landscape paintings,” each frame gently blends into the next—misty peaks dissolve into winding rivers and ancient pines—without any hard cuts. The result is a seamless loop that reimagines traditional brushwork through modern generative AI.
This project began with a question:
​
If AI systems are interpreters of data—translating vast, amorphous sets of information into actionable knowledge—and if this process of “divination” is grounded in an epistemology of continuous optimization, where truth is always provisional and evolving, then what happens when AI tries to reconstruct me?
Also, inspired by Blade Runner—Autoencoded (2016)—a film reconstructed by a neural network that “watched” Blade Runner multiple times and then recreated it from memory—I wondered:
Could I train an AI to become another version of myself? Could a machine simulate my patterns of thought, my face, my mood? And crucially: what is lost, distorted, or erased in that process of machinic translation?
My project explores this through four layers of translation:
-
Myself as a physical body →
-
A 2D image →
-
Numeric data, as interpreted by AI systems (in this case, ChatGPT).And eventually, a fourth translation emerges:
-
My own intervention through editing and re-interpretation, folding the machine’s outputs back into an experimental video.
At its core, my project explores: How is perception distorted, restructured, and reinvented through machinic translations? When AI doesn't “see” images but interprets data, what new, imagined sensory forms emerge? How does the machinic gaze alter our sense of selfhood, embodiment, and presence?
Here, I approach AI systems as contemporary oracles—engines of probabilistic inference, but also sites of epistemological slippage, where the body is both seen and abstracted, both present and lost.
