TL;DR of the paper, generated automatically.

Researchers fit a fluid simulator to a single video to rebuild 3D liquid assets.

• Method: generative 3DGS gives the shape, then a simulator's settings are tuned to match the video's motion.
• Result: assets stay editable, interactive and extendable in time, unlike reconstruction-based methods.
• Scope: the paper says it handles inviscid and viscous liquids, from internet videos.