Diffusion Model — Denoising

flowchart LR
x0[x_0 — image] -->|q add noise| x1[x_1]
x1 --> x2[...]
x2 --> xT[x_T — pure noise]
xT -->|p_θ denoise| x2r[x_T-1]
x2r --> x1r[...]
x1r --> x0r[x_0 — generated]
Net[U-Net ε_θ\nconditioned on t] -.->|predicts noise| x2r
classDef fwd fill:#dbeafe,stroke:#1e3a8a;
classDef rev fill:#dcfce7,stroke:#166534;
class x0,x1,x2,xT fwd
class x2r,x1r,x0r rev