What I'm trying to do is create a cinematic visualization that is fundamentally based on the input of a color image that contains a portrait and that is 670x699 in size. I first create a bunch of random walkers that sample the pixels in that image and convert the colors to gray-scale values of my choosing, which I then use to rebuild the portrait part of the image in gray-scale. So over time you see these random walkers sketch out a face.

That all works great for simply animating the creation of a new PImage. However, just watching a 2D image get built over time is kind of boring, even though it looks very nice. What I really want to do is draw the new face over time using 3D cubes instead. In addition, I want to have a camera that, during the course of the drawing, will move from a regular 2D perspective and swoop down close to and among the cubes in very cinematic way. Finally, I want to animate each of those cubes, moving random ones back and forth a bit along the z-axis, to give the whole project some more life.

Is this idea insane? In my first attempt I simply went for it and tried to represent each of the image's pixels with a corresponding 3D cube, which means I was creating almost 470K cubes and that of course led to a very slow program. I've done some tuning since, trying out versions where I create cubes of edge sizes 3 and 4 to represent 9 and 16 pixels respectively, but I don't love the low resolution image they create. I'm also now using 6 PShapes to underpin all my cubes (6 because I'm using 6 shades of gray) in order to draw in "retained mode" as Daniel Shiffman's PShape tutorial explains.

So, am I insane for trying to create this type of project in Processing? Are there other steps I can take to significantly improve performance or does a project like this require something like Cinder or OpenFrameworks?

Thanks for any advice you can offer! And please let me know if I can provide any other information that would be useful.