AI Termcirca 2020· Added Jun 6, 2026
Diffusion Model
A diffusion model is a probabilistic model used in image generation to iteratively refine images from noise.
Diffusion models are probabilistic generative models that transform noise into coherent data through a series of steps. These models start with random noise and apply a sequence of transformations that gradually improve the data's quality, eventually resulting in a high-quality image. Each step in the diffusion process is guided by learned probabilities that dictate how to adjust the pixels to form realistic images. This iterative refinement makes diffusion models particularly effective for high-fidelity image generation.
Examples
- Generating high-resolution images of landscapes from random noise.
- Creating abstract art by starting with a noise pattern and refining it iteratively.
- Restoring old photographs by applying noise reduction and refinement techniques.
Common misconceptions
- Diffusion models do not operate in a single pass; they require multiple iterations.
- They are not limited to denoising tasks; they can also be used for image synthesis.
- Diffusion models are distinct from GANs but can achieve similar results.
Related terms
Want more like this?
Open the full library
Fresh AI mastery content every 2 hours.