Dream Studio uses machine learning to create images based on sentences and keywords a user enters. Unlike searching for an image based on a keyword in a big data base with tagged pictures, Dream Studio would generate pixels from scratch. Some results are stunning, others have surprising errors (faces that are not finished for example).
You can also mix and match art styles (the image below is a mix up of Mondriaan and Van Gogh).
At the moment these type of services are a gimmick. People try a few creations, share them, and move on. But in a few years from now, this might be the way image “databases” work. No need for that database anymore as photographs are created on the fly,
(Another similar service is DALL-E, but it has a waiting list)