GAN vs Diffusion Model
People often think every AI image maker works the same way. These two do not. A GAN makes a picture in one go, and a second network judges it. A diffusion model starts from noise and cleans it up over many steps. Most image makers today use diffusion.
Generative Adversarial Network (GAN)
Two networks in a contest. One makes fakes, the other tries to catch them.
A generative adversarial network is two neural networks that train by competing. One, the generator, makes fake examples, like fake photos of faces. The other, the discriminator, tries to tell fake from real. Each time one wins, the other learns from it. In the end, the fakes look real.
Diffusion Model
It starts with random dots. Step by step, it cleans them up into a picture.
A diffusion model makes images by removing noise from them, step by step. In training, it takes real images, adds random noise to them, and learns to undo that noise. To make a new image, it starts from pure noise. Then it cleans it up over many steps until a picture appears. A text prompt can steer which picture it makes.
In one line
- Generative Adversarial Network (GAN)
- Two networks in a contest. One makes fakes, the other tries to catch them.
- Diffusion Model
- It starts with random dots. Step by step, it cleans them up into a picture.