Diffusion models are the dominant architecture behind today's AI image generators (Stable Diffusion, Midjourney, DALL-E, Tigmi). They work in two phases. During training, the model is shown millions of images that have noise gradually added until they're pure static — and it learns to reverse that process. At inference, you start from random noise and the model denoises it step by step, guided by your prompt, until a coherent image emerges.
What makes diffusion models useful for interior design is that the denoising can be conditioned on more than text. You can feed in an existing room photo and ask the model to "denoise toward" a different style — that's how img2img works. You can also condition on edge maps, depth maps, or pose estimations using techniques like ControlNet, which is why AI tools can change a room's style without disturbing its layout.
The practical implication: diffusion models are deterministic given a fixed seed. The same prompt + seed produces the same output every time. This is why good AI interior design tools surface a "regenerate" button — you can keep sampling new variations cheaply.