There are signs of memorization of images (or text describing them) from the actual movies. Like in Wall-E and Matrix, you can see, more or less, the robot shape and the green-digits wall in Matrix.
Right. Moreover, it’s visibly memorizing specific images from the movies themselves. It’s very evident for Midsommar and Mad Max, where the renderings contain very idiosyncratic images from these movies which I can’t imagine emerging from mere descriptions of them. So I can only assume that the synopses for these movies contained the actual movie names.