noun
A generative model is an AI system designed to produce new data—such as text, images, or audio—by learning patterns from existing examples. Unlike discriminative models that classify or predict, generative models synthesize content, often using techniques like diffusion models, transformers, or generative adversarial networks (GANs).
These models work by training on large datasets to understand underlying structures. For example, a text-generative model like those behind ChatGPT learns from billions of words to predict the next likely word in a sequence. Similarly, image generators like DALL-E 3 analyze millions of images and captions to create new visuals from text prompts.
For users, generative models change how content is created, reducing the need for manual work but raising concerns about originality, copyright, and ethical use. For instance, a graphic designer might use an AI tool to generate 50 logo drafts in minutes, but they must verify the output doesn’t infringe on existing designs. The cost can be high—training a large model may require millions of dollars in compute resources, though smaller, fine-tuned models can be more affordable.
A marketing team uses a generative model to create 100 variations of a social media ad in seconds, testing which performs best.
In other languages
See also: Diffusion model · Transformer