Computational photography is about to get really good when it can combine hundreds or thousands of frames into one. 1000 frames effectively combined is equivalent to a lens and sensor of 1000x the surface area - ie exceeding a single frame from a DSLR.

Current methods use optical flow and gyroscopes to align images, but I imagine future methods to use AI to understand movement that doesn't work well for optical flow (ie. Where a specular reflection 'moves' on a wine glass).