CoNFiLD-inlet: Synthetic turbulence inflow using generative latent diffusion models with neural fields

Synthetic inflow turbulence generation is a critical bottleneck for high-fidelity, eddy-resolving simulations due to limitations in realism, generalizability, and computational cost. This study introduces CoNFiLD-inlet, a parametric inflow generator that couples conditional neural fields with latent diffusion models to synthesize high-fidelity inflow turbulence with Reynolds number (Re) awareness. Unlike autoregressive or deterministic methods, CoNFiLD-inlet enables mesh-independent, stochastic generation of temporally coherent velocity fields with generalization across unseen Re. Comprehensive validations in DNS and WMLES confirm robustness, scalability, and superior statistical fidelity.