A solver-free approach for modelling stochastic differential equations using conditional normalising flows that enables direct sampling between arbitrary time points within trained horizon.