OK, rather than reading this post you should start by looking at the clip provided by NVIDIA explaining how they have used Artificial Intelligence to let people create images that look like real photos. I think that is just amazing.
So, watch it!
Now, if you paid attention to the presentation you noticed that the researcher presenting the system said “we are not telling the system what a lake, a forest a meadows are, rather the systems has learnt that by looking at thousands of photos and understood the different ways a meadows can look like. It also understood the relation between one component of the image and another, like the different ways a sky may look like if the meadows is grassy or snowy” (wording is mine, expanding on what he actually said, to emphasise the point).
Basically you just start with a sketchy drawing indicating what the various parts are, a waterfall, a tree…. It reminds me of my grandchildren that tell me what the various parts of a drawing are intended to be (actually, that would apply quite well to my chicken scratches…). Then the autonomous image creator re-interpret the whole an proposes its interpretation of it rendering the various parts in a real life way, taking care of their mutual relationship (like placing shadows in the right way, a crucial aspect to fool our brain into believing it is a true photo).
What is really amazing is that all the knowledge on the relationship among various objects has been self learnt by the system just by looking at thousands of other images. By the way, that is exactly how we, during our childhood, learn about the world about us, by looking and experimenting. I guess nobody ever told you that you can only pull a rope, you cannot push it, yet that simply, and complex knowledge, has become part of you.
It is the learning part that fascinates me most. Of course, I can appreciate the implication, the possibilities opened by such a program. Any person can draw a sketch but very few would be able to paint it in a realistic way, probably no one would be able to create a photo quality drawing. Yet, using this system, anyone of us is augmented to become a desk photographer, no need for a camera! People can unleash their creativity in a new dimension, working in the virtual space and obtaining something that looks real. Cool!