A fine-tuned RFdiffusion network enables the de novo design of single-domain antibodies (VHHs) that bind to user-specified epitopes. The study demonstrates that this approach can generate VHHs with high accuracy and specificity, validated through experimental and structural analyses. The RFdiffusion model was trained on antibody structures and fine-tuned to design VHHs with precise binding to target epitopes. The resulting VHHs were experimentally characterized and shown to bind to various disease-relevant targets, including influenza hemagglutinin, respiratory syncytial virus (RSV), SARS-CoV-2 receptor binding domain (RBD), and Clostridium difficile toxin B (TcdB). Cryo-EM structures of the designed VHHs confirmed their atomic-level accuracy, with the VHH backbone and CDR loops closely matching the design model. The study also highlights the potential of RFdiffusion for de novo antibody design, offering a faster and more efficient alternative to traditional methods like animal immunization or library screening. The results suggest that computational design of antibodies can now achieve high accuracy and specificity, with the potential to revolutionize antibody discovery and development. The work underscores the importance of structure-based design and the ability to explore a wide range of CDR loop sequences and structures, which could simplify the optimization of antibody developability features and the targeting of non-immunodominant epitopes. The study also discusses the limitations of current approaches and suggests future directions for improving design accuracy and efficiency.A fine-tuned RFdiffusion network enables the de novo design of single-domain antibodies (VHHs) that bind to user-specified epitopes. The study demonstrates that this approach can generate VHHs with high accuracy and specificity, validated through experimental and structural analyses. The RFdiffusion model was trained on antibody structures and fine-tuned to design VHHs with precise binding to target epitopes. The resulting VHHs were experimentally characterized and shown to bind to various disease-relevant targets, including influenza hemagglutinin, respiratory syncytial virus (RSV), SARS-CoV-2 receptor binding domain (RBD), and Clostridium difficile toxin B (TcdB). Cryo-EM structures of the designed VHHs confirmed their atomic-level accuracy, with the VHH backbone and CDR loops closely matching the design model. The study also highlights the potential of RFdiffusion for de novo antibody design, offering a faster and more efficient alternative to traditional methods like animal immunization or library screening. The results suggest that computational design of antibodies can now achieve high accuracy and specificity, with the potential to revolutionize antibody discovery and development. The work underscores the importance of structure-based design and the ability to explore a wide range of CDR loop sequences and structures, which could simplify the optimization of antibody developability features and the targeting of non-immunodominant epitopes. The study also discusses the limitations of current approaches and suggests future directions for improving design accuracy and efficiency.