It's doing some seriously slow stuff behind the scenes. For example, when the library initialises it creates a 64 * pixelDensity square canvas and then loops through all the pixels in it one by one using a randomly generated transparent color value to draw a 1x1 rectangle to each one (See the 'generateNoise()' function in https://github.com/rikschennink/shiny/blob/master/src/index....). That's just slooooow. HTML5's canvas API has a putImageData method that is much faster because all you're really doing when you fill an image is assigning values to a UINT8 typed array with 4 values per pixel.

It'd be relatively straightforward to achieve this effect using WebGL and a quad to overlay the element, and just pass in the orientation data as a vec2 uniform. That's how some of the effects in my React Neon library work (https://react-neon.ooer.com/) for mouse coordinates. Although, that said, Neon is Chrome-only because it relies on ResizeObserver at the moment. One day I'll fix that...