Generative AI: From GANs to Diffusion Models
How AI creates new images, and the techniques behind modern image generators.
Inspired by learning material from Microsoft
Generative AI creates new content — images, audio, text — that never existed before.
GANs (Generative Adversarial Networks)
Two networks compete: a generator creates fakes, and a discriminator tries to spot them. As they battle, the generator gets remarkably good at producing realistic outputs.
Diffusion models
Today's leading image generators use diffusion. The model learns to start from random noise and gradually 'denoise' it into a coherent image guided by your prompt.
Why diffusion won
- More stable to train than GANs
- Higher quality and diversity
- Excellent at following text prompts
Responsible use
Generative models can create deepfakes and misinformation. Watermarking, provenance, and clear disclosure matter.
Key takeaways
- GANs pit a generator against a discriminator.
- Diffusion models denoise random noise into images.
- Powerful creativity demands responsible use.