It can take both text and/or image as input. You can also choose how coherent you would like results to be. If prompt is descriptive, choose high coherence, whereas if prompt is conceptual and abstract, it's better to choose medium coherence.

Some sample generations from Creative Model

It's really good at working on top of input images.

You can use web app AutoDraw - A helpful AI sketching tool by google to create your sketch and run creative model on it.
Below is an example of the same: