The new generation climate models, so called Convection-Permitting Models (CPMs), provide a step-change in climate simulations by explicitly resolving convection at kilometer-scale resolutions. CPMs run at km-scale spatial resolution and capturing fine-scale surface heterogeneities, thereby improving the representation of extremes. However, their high computational cost limits production of multi-decadal, multi-scenario and large ensemble simulations. This project aims to develop and validate several generative Machine Learning models to emulate Harmonie-Climate (HCLIM) regional climate model simulations at convection-permitting scale (3 km resolution) from coarse resolution inputs (12km resolution). Building on our successful implementation of Super-Resolution Generative Adversarial Networks (SRGANs) for the Emilia-Romagna domain, this initiative will expand research in two critical directions. First, we will evaluate model performance across expanded geographic domains (Southern Sweden and Fenno-Scandinavia) and future climate scenarios. Second, we will implement state-of-the-art Diffusion-Based Generative Models, specifically "Elucidating the Design Space of Diffusion-Based Generative Models" (EDM), to enhance the fidelity and stability of climate emulation at km-scale resolution. The project will rigorously assess the transferability of these models across diverse spatial domains, climate change scenarios, driving Global Circulation Models (GCMs), and temporal scales (historical to future). Given the high-dimensional nature of the HCLIM training data and the iterative sampling requirements of Diffusion Models, the requested GPU resources are essential for efficient model training and large-scale cross-domain validation.