Generating samples from families of related exponential distributions, e.g. distributions with the same sample space, but with fx different temperature or pressure (different 'thermodynamic states), is a challenging problem in the natural sciences and computational statistics. In this project we aim to resolve this issue using a generative AI approach. Specifically, we will leverage the concept of "stochastic interpolants," (Albergo 2022). We will extend this methodology to allow for generalization across thermodynamic states from data only acquired at one or a few states and generalize across states. Stochastic interpolants are a per-sample training strategy to learn transformations between non-normalized probability densities. Once trained, we can transform samples from an easy to sample distribution, e.g. high temperature, to lower temperatures, which are more difficult to simulate. The theoretical framework is an extension of successful approaches such as denoising diffusion probabilistic models and continuous normalizing flows. The outcomes of this research promise an efficient methodology for transforming between thermodynamic states, presenting a significant advancement in computational thermodynamics. Beyond cost reduction, the method allows exploration of physical systems with unprecedented granularity and diversity.