GLOSSARY
Latent Space
The internal numeric space where a model represents concepts — nearby points mean similar things, and moving in a direction means changing the concept smoothly.
Inside every generative model, images, voices and meanings become lists of numbers arranged so that semantic relationships become geometry: king − man + woman ≈ queen was the famous word example, and the same property holds for faces, art styles and product looks. Editing tools exploit this — move a point along a “smile” direction, interpolate between two prompts, or average two images, all without ever naming the concept.
This is why AI tools can do style transfer, image mixing and “make it more dramatic”: those operations are vector arithmetic in latent space. It also explains prompt sensitivity — two nearby wordings can land in noticeably different neighborhoods.