Diffusion models are a family of generative systems that start from random noise and iteratively denoise it to produce images, audio or video. Ho et al.'s 2020 DDPM paper made the architecture practical, and 2022's Stable Diffusion put it on every desktop; since then it has dominated image generation. Compared with GANs, training is more stable and mode coverage is usually better — at the cost of multi-step inference that is comparatively slow. Image Generation, Inpainting, ControlNet and increasingly video systems (Sora, Veo) all sit on diffusion as their underlying architecture.
MEVZU N°124ISTANBULYEAR I — VOL. III
Glossary · Advanced · 2020
Diffusion Models
A family of generative models that produce images, audio or video by iteratively denoising random noise.
- EN — English term
- Diffusion Models
- TR — Turkish term
- Difüzyon Modelleri