It doesn't read any scene data. It only use the viewport render as reference. Cinema 4D itself is irrelevant here, this plugin could use any other 2D image. It gives you a bit more control than only text based instructions.
The issue for 3D artists is that the result is a thematic interpretation and not a true development of the reference scene. This AI reads the image correctly as "a red car on a grey surface with a dark background", but it ignores the shape of the elements, the structure of the scene. The results are interesting, but unpredictable, and not accurate enough for professional works.