Latest Substance Designer experiments - Procedural flora

This project is inspired by a series of documentaries by prof. Jim Al Khalili - Order & Disorder (2012): Exploring the concepts of energy and information.
The documentary is extraordinary, naming countless deep and profound concepts. A few ideas stuck in my mind: - Useful order can be obtained from tapping into the cosmic flow from order to disorder - and - the weaving loom as a precursor to the computer. The punched card/the program, the loom/the hardware, the woven fabric/the output of the program.

I am using similar ideas here but as a simple/toy example - information -> integrator -> output.
The integrator is a dissipative engine. Path = ∫ heading(t) dt is structure-from-sequence — it converts a high-entropy-looking string of angles (which on its own is just a noisy 1D signal) into a low-entropy 2D figure with strong spatial correlations. The information content didn't increase; it got re-expressed in a form humans read as ordered. That's the loom doing thermodynamic work in the Shannon sense. The angle list is the program; the path is the structure that condenses out when the program runs through the operator.
PCA is literally an entropy-sorting operation. What SVD does to your 64 ornaments is rank directions in the data by how much variance they carry, which under a Gaussian assumption is exactly ranking them by information content. The first eigen-mode is the lowest-entropy axis — the most redundancy across samples, the most structure. Truncating at K=8 is choosing how much of the data's order to keep and how much to discard as noise.
Generative procedural systems are dissipative structures with a control surface. When you wire the basis textures into SD and start moving sliders, you're sustaining a flow — energy in, in the form of your attention and the GPU's cycles — and pulling new ornaments out of the latent manifold. Each slider move is a small act of selection against the uniform distribution over possible angle sequences. The sparser and more compact the latent (4–6 sliders instead of 64 raw coefficients, a curve over k instead of independent knobs), the more entropy each gesture extracts per bit of effort. Good UX in a generative system is thermodynamic efficiency — fewer bits of input producing more bits of meaningful output.