Generative AI
Models that learn what data looks like well enough to create new examples: images, text, audio and video.
A normal model is a critic who says 'that is a cat'. A generative model is an artist who has seen so many cats that it can draw a brand new one.
Where it fits
Not shown: 9 more kinds and parts. See the full list below.
+All relationships as a list
Overview
Most ML models are discriminative: they map an input to a label. A generative model instead learns the shape of the data itself, so it can produce new samples that look like they came from the same source.
The main families, roughly in order of appearance:
- Autoencoders compress data and rebuild it; the VAE makes that compressed space smooth enough to sample from.
- GANs pit a generator against a critic until the fakes are convincing.
- Diffusion models learn to remove noise step by step, and turn pure noise into an image. They power Stable Diffusion and most modern image and video generators.
- Autoregressive models generate one piece at a time, each conditioned on the previous ones. GPT-style language models are the famous example.
Everyday use cases
- Case 01
Text-to-image
Type a description, get a picture.
- Case 02
Writing assistants and chatbots
Large language models are generative models of text.