Existing 3D virtual try-on methods suffer from cross-view feature inconsistency, leading to artifacts like perspective-content confusion, view reversal, texture misalignment, and loss of high-frequency details when rendering from arbitrary viewpoints. To address this, we propose AnyView VTON, a novel framework integrating 2D multi-view conditioned diffusion models with 3D Gaussian Splatting. Our approach renders a 3D human model into multi-view 2D images, formulates 3D VTON as a consistent unified 2D VTON process across all views, and reconstructs edited images into a coherent 3D model. Specifically, our approach first performs pose-conditioned feature selection that dynamically chooses frontal/back garments through skeletal analysis while encoding them with camera-parameterized CLIP embeddings for 3D consistent conditioning. Subsequently, cross-view feature fusion resolves garment conflicts through gated fusion blocks while synergistically combining global semantics and local textures via joint attention mechanisms. Evaluated on the THuman2.0 dataset, AnyView VTON outperforms existing methods, achieving superior CLIP directional consistency with garment detail fidelity. User studies show overwhelming preference for AnyView VTON in output quality and clothing alignment. AnyView VTON effectively eliminates cross-view inconsistency, enabling high-fidelity, any-view 3D try-on without expensive 3D scans. This breakthrough enables practical applications in virtual fashion retail, allowing customers to visualize garments from any angle in virtual showrooms.