Understanding image generation using diffusion
Does Figure 10.1 seem normal to you?
Figure 10.1: AI-generated image using diffusion showing the Taj Mahal, India, right next to the Eiffel Tower, France
The Eiffel Tower and the Taj Mahal are situated in two different countries. However, this AI-generated image places them next to each other in a parallel world. This image was generated starting from random noise using a process called diffusion, as shown in Figure 10.2.
Figure 10.2: Generating a realistic image from pure noise using diffusion
Diffusion in the context of generative AI consists of a step-by-step process where simple noise is transformed multiple times to create diverse realistic data, resulting in high-quality sample generation by refining noise iteratively until it resembles the desired data, as demonstrated in Figure 10.3.
Figure 10.3: Diffusion as a step-by-step process of denoising an initial noisy image into a photo-realistic image...
