The original idea about Narrow Band FLIP was published in conference Eurographics 2016, brought by researchers from Technische Universität München, there wasn't any news about implementations of this algorithm in any pipeline tool since then. I tapped into Houdini at the end of last year, and found it a very powerful tool to do almost all kind of things related to computer graphics. So I took some time to implement Narrow Band FLIP into Houdini's own FLIP system with some basic success.

The main idea about NB FLIP is to store particles only on a very thin surface of the whole water body, and driven with previous data to ensure a full FLIP simulation internally.

There are quite a few advantages on NB FLIP:

1. Since very few (precisely, 3-grid-thick) particles are stored and calculated, NB FLIP reduces lots of time on simulation process. The bigger the water body, and the higher the resolution is, the more time it could save. Generally, it can double the calculation speed, in some extreme tests, up to 4x speed improvement is possible compared to original FLIP.

2. NB FLIP can reduce up to 90% or even more particles (check the demo video). Houdini's own Fluid Compress solution can save space required to store particles on disks, but it's a post process. NB FLIP dynamically reduces particles during the simulation, makes it require much less available memory, which means one can boost up the detail resolution up to 10x more particles with limited RAM.

However, since extra calculation process was required during the simulation, those scenes without a whole water body would take somewhat longer time to simulate such as a very shallow stream. Tank-filling style simulation like filling a pool or a bottle of water, the beginning of the NB FLIP simulation is slower than the original FLIP and then boost up later when a decent closed body of water was formed. In this situation, hybrid FLIP is the best solution. Simulating with original FLIP at first and then switch to NB FLIP is possible, thus it could take advantages of both solutions.

Furthermore, NB FLIP supports almost all original FLIP functions, except when one wants to shade the water body's internal movement, which is obviously impossible.

It can also be adapted to A-FLIP and other FLIP variations. Since it's still a developing version which was based on Houdini 15.5, it could be unstable under some extreme conditions, updating to adapt Houdini 16, more optimization and functions are needed to be added.

I hope my study can bring more attention to this for the VFX industry's DCC developers, and let Houdini, Maya, Realflow and other teams make it built-in officially in the future.