Diffusion Models

Best for: Image generation/editing

How it works

Generative models trained to reverse a noising process. Forward diffusion adds Gaussian noise on a variance schedule, $x_t=\sqrt{\bar\alpha_t}\,x_0+\sqrt{1-\bar\alpha_t}\,\epsilon$ over $T$ steps, and a (usually U-Net) network learns to predict the added noise $\epsilon_\theta(x_t,t)$. Training minimises the denoising objective $L=\mathbb{E}_{t,x_0,\epsilon}\bigl\|\epsilon-\epsilon_\theta(x_t,t)\bigr\|^2$. Sampling starts from $x_T\sim\mathcal N(0,I)$ and iteratively denoises back to a clean image, optionally guided by text or other conditions.

Common fields

Design · media · advertising