I agree with Lukas. Even if you had all possible combinations (which you can reproduce anyway, why save them?) you can't just tell the PC to look for "the tentacel monster that rapes her and the spaceship in the background".

The best approach IMO is to search for images on the web, do some shape recognition and try to assemble the desired picture from those.

Another interesting thing you could do is to teach the PC a language, i.e. not only its syntax but its semantics. How cool would it be to tell the PC to search "for that dark colored actress which plays mostly in action flicks and in one movie she's standing in front of a red sports car" and it actually !understands! what you're talking about.

once you can talk to a computer like that everything is possible. even a make game button.

not necessarily. It may understand what an object is and associate attributes to it (like that the actress is dark colored and what an actress basically is) but since it doesn't have any visual input, can't smell, taste, hear, feel it doesn't really know what the object is like for us.

I'm only talking about a complex net of objects with associations to other objects which can simplify a search a lot.