Think of it like sculpting. You start with a block of marble (noise) and then carefully chip away at it (remove noise) until you reveal the final sculpture (image).
Diffusion models are very good, They are used in many popular AI art generators like DALL-E 2 and Stable Diffusion
How do Transformer models generate images if they are not like CNN models? 🤨
Well it doesn't use CNNs for image generation. Instead, It relays on a different kind of model called a diffusion model.
Generating: To create an image, the model starts with random noise. Then, it uses what it learned during training to gradually remove the noise, step by step. With each step, the image becomes clearer and more defined, eventually resulting in a coherent picture.