Diffusion
Also: diffusion model, latent diffusion
The generative technique behind most local image, video, and audio models. It starts from random noise and removes it step by step, guided by your prompt, to arrive at an image, clip, or sound.
Where a language model predicts the next token, a diffusion model works by denoising. It starts from random noise and, over a series of steps, gradually removes it to reveal an output that matches your prompt. The number of steps trades speed against quality, and is one of the main dials in an image or video tool.
Almost every local image model, such as SDXL and FLUX, and the newer video and music models, is a diffusion model. This is also why their hardware needs are driven by resolution, clip length, and step count rather than by context, and why tools like ComfyUI expose the sampler and step count directly.