Reproduce “Fluid-Gen-Zero: Grounding Pretrained Video Generators in Physics without Training” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hong Huang, Yuqiu Liu, Chenyu You, Daniel Martin, Chuhang Zou, Wuyang Chen
Fluid-Gen-Zero guides frozen video generators with simulated fluid-object motion, without updating their parameters. It reconstructs scenes for simulation and injects projected motion into denoising, using different guidance for object and fluid regions. Across 43 benchmark samples, it improves alignment on three backbones; on Tora, trajectory error falls from 133.20 to 24.61 pixels. The reference is simulated, so alignment depends on reconstruction and simulation accuracy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.