Determining the three-dimensional (3D) structure of RNA is crucial for understanding its diverse biological functions. The field of computational RNA structure prediction has recently been transformed by deep learning, which has dramatically improved accuracy beyond that of conventional homology- and de novo modeling approaches. This article overviews these advancements. We first summarize the principles of foundational conventional approaches before detailing the current state-of-the-art deep-learning-based approaches. Deep-learning-based approaches are categorized into strategies that leverage multiple sequence alignments (MSAs), recent MSA-free methods that rely on single sequences, and emerging generalist models that can predict entire heterogenic biomolecular complexes. Furthermore, we discuss how these predictive breakthroughs are accelerating the field of RNA design. Finally, we outline the current challenges and future directions for computational RNA structural biology.