Skip to content
AI Atlas

Generative AI

From judging what something is to synthesising what it should be

OVERVIEW

All Concepts
6
Beginner
2
Intermediate
3
Expert
1

A generative model is not content with labelling a sample; it must produce the sample itself. Autoencoders, GANs, diffusion models and autoregressive models are four principal routes, each giving a precise mathematical formulation to the vague goal of “looking real” — and each now underpinning image, video, speech and 3D content production.

Questions this domain answers

  • Q1

    What distribution is a generative model learning?

  • Q2

    Why is diffusion more stable than a GAN?

  • Q3

    How does a text prompt become an image?

Concepts in this domain

  1. 01Generative Models: An OverviewBeginnerDiscriminative models answer "what is this"; generative models answer "what should this look like"
  2. 02Autoencoders & VAEIntermediateSqueeze information through a bottleneck, then let it grow back
  3. 03Generative Adversarial NetworksBeginnerA forger versus an inspector: each pushes the other to its limit
  4. 04Diffusion ModelsIntermediateLearn a thousand tiny denoising steps, and you can build an image from pure noise
  5. 05Latent Diffusion & Conditional ControlExpertRun diffusion not over pixels, but inside a compressed semantic space
  6. 06Multimodal GenerationIntermediateOne model that learns to speak, to draw, to move — even to model the 3D world