AI Glossary / Diffusion
🎨 Image, video and audio
Diffusion
An image generation technique: the model "denoises" random noise until an image emerges. Powers Midjourney and Stable Diffusion.
Diffusion models generate images by reversing a destruction process: during training they learn how an image gradually turns into random noise, and during generation they walk that path backward - starting from pure noise and step by step "denoising" it until an image matching the prompt emerges. This technique powers most modern image generators, and its variants also drive video and audio generation. Practical intuition: more denoising steps usually means more detail, but slower generation.
Related concepts