Diffusion model
확산 모델(디퓨전)
An AI method that starts from static-like noise and polishes it step by step into an image or a video.
Overview
A diffusion model first watches, over and over, as a perfectly good picture is slowly mixed with noise until it is completely ruined. In doing so, it learns the reverse: how to remove the noise and bring the picture back.
To make an image, it starts from meaningless noise and removes noise many times to fit the description you asked for. A clear picture gradually appears. AI that makes video and music often uses this method too.
An everyday example
It is a little like wiping a foggy window bit by bit until the scene outside appears. The difference is that the scene is invented to match your description.
Why it matters
Most image and video generation AI comes from this method. As it became easy to make fake images that look like photos, telling real from fake has become important.