Repurposes a 2B-parameter latent video transformer as a differentiable physics simulator for urban wind flow optimization.
March 24, 2026
Original Paper
Pretrained Video Models as Differentiable Physics Simulators for Urban Wind Flows
arXiv · 2603.21210
The Takeaway
It treats a pretrained video model as a surrogate for complex CFD simulations. Because the model is differentiable, building layouts can be optimized directly via backpropagation through the 'physics' hallucinated by the video model, enabling 1000x faster design exploration.
From the abstract
Designing urban spaces that provide pedestrian wind comfort and safety requires time-resolved Computational Fluid Dynamics (CFD) simulations, but their current computational cost makes extensive design exploration impractical. We introduce WinDiNet (Wind Diffusion Network), a pretrained video diffusion model that is repurposed as a fast, differentiable surrogate for this task. Starting from LTX-Video, a 2B-parameter latent video transformer, we fine-tune on 10,000 2D incompressible CFD simulatio