Search⌘ K
AI Features

Inference Strategies for Diffusion Models

Explore how diffusion models generate images through iterative denoising and why naive inference can be slow. Understand the inference strategies DDIM and DPM-Solver that accelerate generation by skipping or approximating denoising steps without retraining, balancing speed and output quality for real-world applications.

Diffusion models generate images by gradually transforming noise into structure. At inference time, the model starts with a completely noisy image and repeatedly applies a denoising network to remove noise step by step. Each step produces a slightly cleaner image than the previous one.

This process works well, but it is inherently slow.

What happens during diffusion inference

Consider a diffusion model trained with ...

The diffusion inference process
The diffusion inference process

Each of these steps requires a forward pass through a large neural network. If a single forward pass takes, for example, 20 ...