Latent space
Also: latent representation, latent code
A latent space is the set of internal vectors (latent variables) that a model uses to represent data, usually far lower-dimensional than the raw input. In a well-trained generative model, moving smoothly through it changes the decoded output smoothly.
Example
In a 2-D latent space learned from handwritten digits, all the 1s cluster together, and decoding points on the line between a 3 and an 8 morphs one digit into the other.