Very cool. I'm surprised that it is so hard to stitch the frames together, and requires acceleration data. Naive I guess, but I was imagining you could simply examine the frames and work out how to join them nicely in code.

It'd be cool to see some ridiculously wide, scrollable images from this process.