How does AI make something real?

I fed different kinds of drawings to a Runway ML model that tries to turn drawings into realistic photos. This was a research in how an AI views a drawing and what it makes of it, and what kinds of drawings it was able to interpret best.

First I tried linear shapes.

These images all turned out very similar, and I was a little disappointed. I thought maybe it would work better if I added in gradients, so I tried that next.

This indeed worked way better. I was especially curious about that last image. I was wondering how different these colours would be if I used a series of moving gradients, would they all be the same colours or would they be randomised? I then made a couple of these, and lined up the images in a GIF.

Two quick experiments:

And the final GIF: