This breathtaking imagery is neither generative AI nor computer graphics—but rather real physical liquids under a microscope, exhibiting "reaction-diffusion" behavior. These clips were created and filmed by artist Kamil Czapiga: (ig: cosmodernism)

However, there is an AI connection: These so-called "Turing patterns" were first studied by Alan Turing, who in addition to being a pioneer of Computer Science & Artificial Intelligence, developed a mathematical description of how animals get their spots:

The principle is surprisingly simple: Two chemicals spread out across a surface ("diffusion"), and the relative concentration at each point determines how they interact ("reaction"). (In ML terms, one might say the former is a linear equation; the latter is a "nonlinearity.")

These equations have been numerically simulated many times, to great effect—for instance in Greg Turk's 1991 SIGGRAPH paper "Generating Textures on Arbitrary Surfaces Using Reaction-Diffusion":

You can find numerous versions on Shadertoy, like these beautiful multiscale patterns: (I would say "seems ripe for a neural renaissance," but I'm sure several papers have appeared on arXiv in the time it took me to write these posts!) Enjoy. [n/n]

I strongly suspect most of the pattern formation seen here is /not/ reaction diffusion... and that these are not Turing patterns, ferrofluids have a different phase space.

It’s an interesting question. Many systems with very different dynamical origins exhibit the same limiting behavior. E.g., Allen-Cahn (reaction diffusion) agrees with mean curvature flow (MCF) in the limit. Here also you have surface tension (MCF) and nonlocal forces (repulsion).

This is why the special effects of the nebula/star in The Fountain (2006) are so timeless:

Wonderful film.

Oddly similar to ocular dominance columns? 🤔

Not odd at all—reaction-diffusion is a basic mechanism determining cell differentiation/pattern formation in chemistry and biology.