When I obtained my doctoral degree in 2019, I worked primarily as an image-processing operator in design, handling various image-processing tasks every day. At that time, the field was constantly discussing both the marked progress of the Diffusion Process proposed by Jascha Sohl-Dickstein and others and the rapid democratization of image-processing technology.*1

In late 2021, a sudden episode of phantom pain (an abrupt pain or impact felt at an old injury site) prompted me to begin restoring the record photographs I had accumulated and to embark on what I call “editing.” The phantom pain was a cry of the body, beyond the control of my reason, and my body itself led me to start the present project.

​

My purpose is not to speak out as a victim but, as a maker, to sublimate a trauma that behaves like a wild horse into the work. By meticulously pursuing the phenomenon of violence—peeling away memories and scabs one photograph at a time—I continue to question the whereabouts and transformation of a body corroded by violence. This project will become a lifelong undertaking.

In this work, “editing,” understood as Somatic excavation, is a technical metaphor within my collage-based workflow; it refers to the entire act of integrating fragmented experiences and memories inside a local dataset space—namely, a personal data repository that exists in the feature space of personal data. Such an approach would never have been possible without the remarkable development of these technologies and their open democratization.

​

The series Denoise Body Experiments relies on multiple stages, including photography, the application of a latent diffusion model to the recorded images, several image-processing filters written as .py scripts, editing in Photoshop, and printing.

The reverse diffusion process, one of the theoretical frameworks for diffusion models (reverse SDE: dx = [f(x,t) − g²(t) ∇ₓ log pₜ(x)] dt + g(t) dw), paradoxically learns generation methods by gradually destroying data. Beginning from complete noise, it extracts structure and reconstructs the signal step by step; the resulting images are samples drawn from probability distributions.

From the artist’s standpoint, this protocol of generation through gradual destruction is conceptually continuous with stitching back together, by Bayesian inference, the body conceived as information dissipated by trauma. Although the early stages of the project resembled a crude therapeutic dissection, trial and error seamlessly linked the creative approach to the engineering method. Just as the diffusion model extracts patterns that humans perceive as meaningful structures according to probability distributions, the work creates a new wholeness from a body fragmented after injury—an editing scarcely possible in the physical space in which we live.

​

The emergence and fragmentation of image groups through latent diffusion and other methods correspond to the attitude of casting the body into latent space as a single grain of noise. This approach enables multiple temporal states to be integrated onto a single visual plane and edited in digital space. In this way, sensibility and engineering methods harmonize without contradiction. Collage also serves as a technical metaphor for manipulating nature and society. The work actively offers the body to, and retrieves it from, that editing process. What returns from the far side of the diffusion process is a coded body; even though it belongs to the same model, it is already supported by a different axis. The work engages in dialogue with that body. In this sense, the project is a correspondence awaiting a reply, a resonance close to prayer.

​

For the record photographs and reference material, I built a local dataset of about 1,000 personal images to serve as training data for the latent diffusion model. The model was implemented on a standard PyTorch framework, using a U-Net architecture to estimate either the mean μθ or the noise ϵθ of the diffusion process. For image processing, I primarily used batch filters written in OpenCV to perform color adjustment, blurring, and sharpening.*3 NumPy optimized computation. All stages were carried out through a command-line interface, practicing impromptu creative coding—closer to writing poetry than to engineering efficiency, like the moment of lifting a freshly printed woodcut sheet.

Once the photographic base is complete, the images are printed as giclée prints, hand-colored with dye ink, drawn upon, and framed. After production, the local dataset is discarded.

​

​

​

 

References:

*Note1: 
[1] Sohl-Dickstein, J., Weiss, E. A., Maheswaranathan, N., & Ganguli, S. (2015). Deep unsupervised learning using nonequilibrium thermodynamics. International Conference on Machine Learning.

[2] Song, Y., Sohl-Dickstein, J., Kingma, D. P., Kumar, A., Ermon, S., & Poole, B. (2021). Score-Based Generative Modeling through Stochastic Differential Equations. ICLR.

Both references retrieved from Google Scholar, August 11, 2024.

​

*Note2:
This phantom pain in 2021 was an extremely physical bodily response stemming from experiences of violence (which I had been pretending never happened) during my childhood through adolescence. I was surprised because I couldn't control it with my own will. This experience provided me with a specific creative objective: the aesthetic sublimation of trauma.

 

*Note3: 
However, to visualize progress, the tqdm library was utilized to track the waiting time during filter processing in the .py script.