Nanobodies (VHHs, ~12-15 kDa) are single-domain antibody fragments valued for diagnostic applications, yet their engineering is frequently limited by poor solubility and aggregation. This study used the aggregation-prone anti-VirB4 nanobody a_b5 as a model to establish a two-stage solubility optimization strategy: rational single-point mutagenesis, then AI-driven de novo design. Consensus sequence analysis and CamSol prediction converged on glycine 54; G54A and G54D mutations reduced the Z-average hydrodynamic diameter by ~31% and ~32%, respectively, but particle sizes remained in the hundreds of nanometers, revealing an inherent ceiling for single-point mutagenesis. To surpass this limit, a modular AI pipeline integrating AlphaFold3 hotspot analysis, RFdiffusion backbone generation, dual-pathway ProteinMPNN sequence design (Vanilla and Soluble), AlphaFold3 ipTM screening, and 100-ns molecular dynamics was constructed. Among 780 designed sequences across 10 nanobody-antigen pairs, B5S achieved a CamSol solubility score of 3.007 (480% increase over wild-type 0.518). DLS confirmed a Z-average diameter of 6.084 nm with 43.4% monomer peak; B5S was solubly expressed in E. coli while the Vanilla counterpart B5V precipitated. Direct ELISA confirmed dose-dependent binding of B5S to VirB4 with apparent K_D ~856.2 nmol/L. These results demonstrate that AI-driven holistic redesign can simultaneously optimize binding interface quality and macromolecular solubility.