Can we predict if DALL-E 2 will create a good image or not for a given text?

I have done a small experiment to see if we can predict when DALL-E 2 will generate a good image. Details about the experiment are at https://twitter.com/nutanc/status/1521906864589393920 Tl;Dr; I think by mixing and matching we can see when Dalle 2 works. 1. Test sentence embedding on a QnA model. 2. Search Google for images on that sentence. If both 1 and 2 work then Dalle2 will give a good result.