A Game Freak programmer has teamed up with a University of Tokyo professor to develop a new technique that makes complex water simulations faster while preserving much of their visual quality. The research could help improve the efficiency of high-resolution liquid simulations, particularly in scenes featuring detailed waves and rainfall.
Fumiya Narita of Game Freak and Takashi Kanai, a professor at the University of Tokyo, presented their paper, “Beyond Horizontal Refinement: Extending Quadtree Tall Cells with Adaptive Optical Layers,” at Pacific Graphics 2026, an international computer graphics conference held in Singapore on October 6.
The new technique, called Extended Quadtree Tall Cells, builds on two existing approaches to liquid simulation. These methods reduce computational demands by adjusting how water is represented in a simulation, allowing greater detail to be preserved near the surface without requiring the same level of precision throughout the entire volume.
Previous techniques had limitations when dealing with extensive water surfaces covered in small-scale details and rapid fluctuations. The new approach combines adaptive horizontal grid refinement with adjustments to the thickness of the high-resolution layer near the water’s surface, allowing it to reduce computational demands in scenes where earlier methods were less effective.
In testing, the researchers found that the technique could deliver significant improvements in certain scenarios without noticeably compromising visual quality. Across three test scenes, including one featuring rainfall and numerous small splashes, the method achieved pressure calculation speeds approximately 2.5 to 3 times faster than Quadtree Tall Cells alone. Overall simulation performance improved by around 1.5 times.
The benefits were not universal, however. Scenes that were already efficiently handled by the previous technique, such as certain dam-break simulations involving pillars and slopes, saw smaller gains. The researchers also identified cases in which the new method could take slightly longer to process, while some shallow-water scenarios produced minor visual differences. Its current implementation also consumes more memory despite reducing the number of grid cells used in calculations.
Further work will focus on reducing the additional computational overhead involved in constructing the adaptive layers and improving the handling of interactions between liquids and more complex solid objects.
Game Freak is best known as the developer of the Pokémon series, making the research particularly interesting given the franchise’s emphasis on exploring expansive environments. However, the new technique has not been demonstrated as a real-time water simulation system for games. The experiments involve high-resolution simulations that can take more than ten seconds per step, rather than the real-time processing expected during gameplay.
Consequently, while the research demonstrates a promising way to make detailed liquid simulations more efficient, it should not be taken as confirmation that an upcoming Pokémon game or another Game Freak project will use the technology. Its potential applications in game development remain to be explored.