A method for simulating fluid surfaces in real-time in response to user input includes detecting interactive conditions triggering insertion of a heterogeneous mesh sequence in a 3D model sequence for rendering, fetching ones of the heterogenous mesh sequence from a computer memory, inserting the successive members in corresponding representations of the 3D model sequence in a computer memory, and rendering successive video frames from the representations of the 3D model sequence each including a successive member of the heterogenous mesh sequence. A related method for generating a compact heterogeneous mesh sequence for use in rendering corresponding frames of video includes generating a heterogenous mesh sequence modeling response of a fluid surface to physical forces, the heterogenous mesh sequence characterized by position values represented in computer memory by not less than 12 bytes for each vertex thereof, transforming the heterogenous mesh sequence into the compact heterogeneous mesh sequence, at least in part by quantizing the position values to not greater than four bytes, and storing the compact heterogeneous mesh sequence in a computer memory for use in real-time rendering.