1. The Physics of Textile Failure in Latent Space

When text-to-image diffusion models synthesize fabrics, their latent self-attention layers naturally aggregate pixel values toward low-entropy, smooth gradients. The resulting textile looks visually plausible from a distance but fails instantly upon close inspection: cashmere turns into airbrushed plastic, crushed velvet loses its multidirectional directional sheen, and raw silk misses the natural slub imperfections that define organic luxury.

To achieve high-frequency textile tactile fidelity, prompt architecture must emulate optical physics: anisotropic surface reflectance, subsurface scattering across individual fiber threads, and cross-directional thread micro-tension.

2. The Three Sovereign Textile Blueprints

Blueprint A: Uncarded Mongolian Cashmere

Cashmere fibers have an average diameter of 14–19 microns with high micro-crimp, giving them loft and a soft halo under glancing light. Do not use generic words like “cashmere sweater”. Use token syntax that defines the fiber morphology:

Blueprint B: Raw Habotai & Slub Silk

Real raw silk contains uneven thread diameters (slubs) that create irregular specular breaks. We induce this by forcing the model to calculate microscopic irregularities:

Blueprint C: Heavy Crushed Velvet (Directional Pile)

Crushed velvet requires opposing specular reflections: areas where the pile faces the light reflect brightly, while inverted pile creates profound, saturated shadow wells.

3. Lighting Setup: Grazing vs. Frontal Diffusion

Frontal softbox lighting flattens fiber pile by eliminating the microscopic cast shadows between thread ridges. For gallery-grade textile fidelity, always prescribe grazing cross-light: