Back to timeline

Denoising Diffusion Probabilistic Models

Ho, Jain, and Abbeel publish DDPM, helping establish diffusion as a leading approach to high-quality image generation.

Research

What Happened

In June 2020, “Denoising Diffusion Probabilistic Models” (DDPM) was published on arXiv, presenting strong image synthesis results with diffusion-style generative modeling.

Why It Matters

Diffusion methods later became central to modern text-to-image systems and creative tooling, reshaping the generative AI landscape alongside large-scale transformers.

Technical Details

DDPM frames generation as iterative denoising: a model learns to reverse a gradual noising process, enabling high-fidelity sampling with iterative refinement.