Molecular dynamics (MD) simulations are indispensable for studying the dynamic behavior of biomolecules, yet they are often constrained by timescale challenges that limit sampling efficiency. Generative machine learning (ML) models have recently emerged as promising tools to address these limitations. In this study, we revisit the potential of the Denoising Diffusion Probabilistic Model (DDPM) as an approach for enhancing the sampling and for generation of atomistically accurate conformational ensembles of biomolecules of diverse size, ranging from small folded proteins to large intrinsically disordered proteins (IDPs) and intrinsically disordered regions (IDRs). By training DDPM independently on MD simulation data of torsion angle and all-atom coordinates, we demonstrate their ability to effectively explore conformational space, including sparsely sampled regions, and to generate novel conformations. The performance of DDPM is evaluated on a diverse set of systems, including the Trp-cage mini-protein, the folded protein BPTI, the IDP -Synuclein, and the IDR Ash1. Our results reveal that DDPM-generated ensembles accurately reproduce key structural features such as secondary structure, radius of gyration, and contact maps, while also capturing the inherent conformational heterogeneity of IDPs. These findings highlight the potential of DDPM to enhance the efficiency and accuracy of biomolecular simulations, offering a scalable framework for gaining deeper insights into protein structure and dynamics.