This application is designed as a demonstration of the underlying idea.
No caching is done, at all, each time it moves it redraws the image by recreating the entire field. The actual implementation uses no hardware acceleration, though the algorithm was designed specifically with GPU hardware in mind. However, these limitations are by design, because again, it's a demonstration.
On top of that it also functions as a useful tool. If you need to produce images of random noise on your phone or tablet for some reason.
The noise algorithm was originally based on Diamond Squared by Fournier, Fussell and Carpenter. And a need to surpass the given limitations on size, memory, and usability.
This was achieved primarily by ditching the base case of a 2x2 toroid and going with an infinite field of perfectly random values, and then developing a scoping algorithm which allows for any number of repeated convolutions within infinite regions. Getting rid of the the memory limitations I developed methods for performing all image convolutions within their own memory footprint (independently wildly useful). Resulting in an algorithm which requires no memory (it does produce 2 extra rows and columns of garbage from the blur stage, which is cut off).
Allows for the production and saving of image noise regions directly from your tablet or phone.
If you've seen one bit of it, you've seen it all. Because it's fractal it basically all alike (and dissimilar) and goes on forever.