Is the Elephant Flying? Resolving Ambiguities in Text-to-Image Generative Models