AI Foundationsbeginner12 min · 9 steps
Image Generation: GANs and Diffusion
How networks create brand-new images — from the adversarial game of GANs to the step-by-step denoising behind Stable Diffusion and DALL-E.
datarequestresponsesignal
Key ideaStep 1 of 9
Image generation: create pixels that never existed
Image generation creates new, realistic images from scratch. Unconditional generation starts from random noise alone; conditional generation also takes a condition, most often a text prompt (text-to-image).
Most generators work in a latent space: a lower-dimensional space of codes where nearby points decode to similar images. Two families dominate the field: GANs and diffusion models.
InsightUses: art and illustration, design assistance, virtual worlds and synthetic training data (data augmentation).
What changed
- +Noise in
- +Optional text condition
- +Image out
← Previous lessonVision Models: CNNs, Classification and DetectionNext lesson →RAG and Vector Stores
Lesson 6 of 34 in AI Foundations