Researchers developed a machine learning model that understands the underlying relationships between objects in a scene and can generate accurate images of scenes from text descriptions.