Designing antibody sequences to better resemble those observed in natural human repertoires is a key challenge in biologics development. To this end, we introduce IgCraft: a multi-purpose model for paired human antibody sequence generation, built on Bayesian Flow Networks. IgCraft achieves competitive or state-of-the-art performance and reasonable efficiency on a broad range of antibody sequence generation tasks, including unconditional sampling, sequence inpainting, inverse folding, and CDR grafting for humanization. By unifying these tasks within a single generative model, IgCraft provides a scalable framework addressing a variety of important challenges in computational antibody discovery and engineering.