A diffusion model is a type of generative AI that creates images by starting from random noise and removing it step by step until an image appears that matches a description. It learns to do this by being trained the other way round: adding noise to millions of images and learning to undo it.
Why it is called “diffusion”
The training process gradually “diffuses” an image into noise, like ink spreading in water. The model learns the reverse path. At generation time, a text prompt steers each denoising step towards what you asked for.
Where you meet it
Most current image tools are built on diffusion models. A key step was latent diffusion (Rombach et al., 2022), which runs the process on a compressed version of the image and made high-resolution generation practical; it is the basis of Stable Diffusion.
In interior design
On its own, a diffusion model invents a room. To redesign your room, tools start from your photo (img2img) and constrain the result with structure from it (ControlNet), so the new image keeps your walls and windows while the furniture and finishes change.