In this talk, I will explore the application of generative models, specifically neural importance sampling (NIS), for efficiently computing phase space integrals in high-energy physics. Traditional Monte Carlo methods, such as the VEGAS algorithm, face challenges in accurately handling complex and expensive calculations required for modern experiments like those conducted at the LHC. This work focuses on overcoming these by introducing ZüNIS, a fully automated NIS library tailored for HEP applications but also applicable beyond it. I discuss extensions to the NIS formulation to improve stability and performance, along with the user-friendly design of ZüNIS, making it accessible to non-experts. Benchmark results on both toy and physics examples demonstrate both the effectiveness of NIS realised by ZüNIS and other implementations, as well as the limitations of black box approaches. The talk highlights the potential of generative models to revolutionize phase space integrals computation in HEP, paving the way for more efficient simulations for the high luminosity era.
