How pictures are produced from descriptions.
HOW IT WORKS BROADLY
Models trained on very large collections of images with text descriptions.
They learned the relationship between words and visual features, and generate new images matching a description.
WHAT IT IS GOOD AT
Illustrations and concepts Backgrounds and textures Variations on a theme Visualising an idea quickly
WHAT IT IS BAD AT
Text within images, frequently garbled Precise anatomy, particularly hands Consistency across multiple images Accurate representation of specific real things
THE PRACTICAL USES
Concept visualisation before commissioning real work Placeholder imagery Illustration where photography is impractical
WHAT TO BE CAREFUL ABOUT
Copyright. The legal position on training data and on generated output is contested and varies by jurisdiction. Representing real people, which raises separate concerns. Using generated images as though they were photographs of real things.
FOR BUSINESS USE
Your own photographs remain more persuasive than any generated image, particularly for a local business.
Generated imagery suits concepts, not evidence.
