LATENT
HOW AI WORKS - FRAME BOOK
VOL. 01·SEP 2026

01DIFFUSION

FROM NOISE TO STRUCTURE.

A PROGRESSIVE REFINEMENT PROCESS THAT TRANSFORMS RANDOMNESS INTO ORGANIZED GEOMETRY THROUGH ITERATIVE DENOISING.

  1. 01NOISE

    x0∼N(0,I)x_0 \sim \mathcal{N}(0, I)

    Random pointsin latent space.

  2. 02COARSE FORM

    x1=x0+ϵ⋅∇θ(x0)x_1 = x_0 + \epsilon \cdot \nabla_\theta(x_0)

    Local structureemerges.

  3. 03CONTOURS

    x2=fθ(x1)x_2 = f_\theta(x_1)

    Contours alignand stabilize.

  4. 04GEOMETRY

    x3=fθ(x2)x_3 = f_\theta(x_2)

    Clear geometricform appears.

  5. 05REFINED

    x4=fθ(x3)x_4 = f_\theta(x_3)

    Structured shapein high fidelity.

ITERATION (tt)

ONION SKIN(TEMPORAL)

ONION SKIN (TEMPORAL)

02MORPH TRANSITIONS

CONTINUOUS TRANSFORMATIONS THAT CONNECT VISUAL CONCEPTS ACROSS THE SERIES.

A UNIFIED FLOW OF GEOMETRY WHERE ONE FORM EVOLVES INTO THE NEXT.

  1. 1GRID→POINT CLOUD

    G(u,v,w)G(u, v, w)

    Regular lattice.

  2. 2POINT CLOUD→RINGS

    P={pi∈R3}P = \{p_i \in \mathbb{R}^3\}

    Sampled points.

  3. 3RINGS→PLANES

    R(r,θ)R(r, \theta)

    Circular structure.

  4. 4PLANES (LAYERED)

    L={Pk∣k=1…n}L = \{P_k \mid k = 1 \ldots n\}

    Layered composition.

Diffusion &Morph Transitions

From Noise to Form, Across Concepts.

TWO CONNECTED SYSTEMS THAT ILLUSTRATE HOW INTELLIGENCE EMERGES FROM RANDOMNESS AND HOW VISUAL IDEAS TRANSFORM CONTINUOUSLY THROUGH GEOMETRIC OPERATIONS.

DIFFUSION
x0→xt→xx_0 \to x_t \to x
MORPH TRANSITION
A→BA \to B
f(A,t)=(1−t)A+tBf(A, t) = (1 - t)A + tB
CONTINUOUS SPACE
z∈Rnz \in \mathbb{R}^n
SHARED LATENT
zt=g(z,t)z_t = g(z, t)
FORM EVOLUTION
∂x∂t=F(x,t)\displaystyle \frac{\partial x}{\partial t} = F(x, t)