LARRUZ explains how fluid simulations lose fine detail when they divide space into chunks to make computing feasible. Wavelet turbulence synthesizes the missing high-frequency details cheaply and won a Technical Oscar. The technique is now built into Blender.
How can we simulate the motion of fluids and smoke? If we had a block of plastic in our computer program, and we would add the laws of physics that control the motion of fluids, it would immediately start behaving like water. In these simulations, we're mostly interested in the velocity and the pressure of the fluid, how these quantities exactly change in time. This we need to compute in every point in space, which would take an infinite amount of resources. What we usually do is we try to compute them not everywhere, but in many different places. And we try to guess these quantities between these points. By this guessing, a lot of information is lost, and it still takes a lot of resources for a really detailed simulation. It is not uncommon that one has to wait for days to get only a few seconds of video footage. And this is where wavelet turbulence comes into play. We know exactly what frequencies are lost and where they are lost. And this technique enables us to synthesize this information and add it back very cheaply. This way one can get really detailed simulations at a very reasonable cost. Here are some examples of smoke simulations with and without wavelet turbulence. It really makes a great difference. It is no accident that the technique won a Technical Oscar award among many other systems. It is implemented in Blender, so anyone can give it a try. Make sure to do so because it's lots of fun. The paper and the supplementary video is also available in the description box. This is an amazing paper, easily one of my favorites. So if you know some math, make sure to take a look. And if you don't, just enjoy the footage. Thank you for watching and see you next time.
