FuncGenFoil is a new generative model that designs and edits high-fidelity airfoils as continuous functions, outperforming prior methods in accuracy and diversity while enabling flexible, resolution-free shape generation.

Inference-time scaling for flow models

A rectified flow-based generative model that improves generation quality and straightens ODE solution trajectories by incorporating real images via slerp into the reflow training process.

Energy-based training of neural continuous-time Markov processes in general state space.