So... what the ai does?

Because it learns what a concept looks like and that's how it will generate an image from pure static. It doesn't have images saved inside the model it has knowledge of what something looks like and how to make that from static. Essentially it knows how to draw a thing but doesn't have images of the thing saved in its data.

https://www.fastcompany.com/90649913/the-unusual-creative-process-of-the-artist-behind-the-little-mermaid-and-beauty-and-the-beast

Aphantasia prevents the generation of mental images based on knowledge of what things look like, but it does not prevent that knowledge serving as the basis for an image made with pencil and paper. Keane can draw a picture of Ariel because he knows what humans (and fish) look like, and that information—plus the skills acquired through study and practice—steers his hand accordingly.

This is the best explanation i can think of for this.