Diffusion Model

This painting visualizes the process of a Diffusion Model, from entering a prompt to the generation of an image. As the scene progresses from left to right, noise is gradually removed, and the image takes on increasingly meaningful form. At the center, a circular motif represents the state of “in progress” — a loading circle, the familiar spinning icon shown while something is being generated.

The composition is inspired by Tōhaku Hasegawa’s Pine Trees Folding Screens, reinterpreting the guided flow of the viewer’s gaze as distortions of pixels. The traditional Japanese concept of ma (the interval, or space between things) is here layered onto the mechanism of generative AI and reimagined in a contemporary context.

Just as Pine Trees Folding Screens is composed as a pair, this work is also presented together with its counterpart, “GAN”, which depicts another generative model.

These paintings are about Ma (間). Ma is a Japanese sense of both time and space — often understood as a meaningful pause or an empty gap. Traditionally, Ma can be seen in art such as Hasegawa Tōhaku’s Pine Trees Screens, where the empty is not nothing but a meaningful presence.

Today, Ma can be found in the process of image generation by AI. For example, when we see “Loading” or “Creating,” it is the time when we are waiting for calculation. The “loading” or “creating” screen may look like wasted time. But in Japanese culture, people have found rich meaning and beauty in such empty intervals. This work reinterprets that sensibility for the digital age.