Diffusion

Diffusion is a core technique behind many generative AI models for art, music, and text. It works by:

• Gradually adding noise to data (like images or audio) until the original structure is destroyed.
• Training an AI model to reverse this process: starting from noise and step-by-step removing it to reconstruct meaningful data.

Unlike physical diffusion (which is irreversible), AI diffusion models learn a reversible “undo” process that lets them generate new images, sounds, or text from pure noise.