In the structural design of thin-walled structures, generating complex 3D geometries that satisfy specific design parameters is a fundamental challenge. Previous studies on parameter-to-3D generation tasks have predominantly utilized implicit representations such as Signed Distance Functions (SDF). However, these conventional methods are fundamentally limited to watertight, closed volumes and cannot represent zero-thickness surfaces (open surfaces), which are critical for modeling ideal thin plates in engineering applications. This study addresses this limitation by investigating a generative framework specifically designed for open-surface structures. We apply a state-of-the-art latent diffusion model, Surf-D, which employs Unsigned Distance Fields (UDF) to handle arbitrary topologies including open manifolds. We adapt this architecture to condition the generation process on explicit design parameters, enabling the direct inference of 3D thin-walled shapes from low-dimensional inputs. The objective of this research is to verify the feasibility of generating high-fidelity, thin-walled models that accurately reflect input geometric constraints, extending the capabilities of deep learning-based design support systems. We evaluate the method's performance in reconstructing and interpolating thin-shell geometries, demonstrating its potential to overcome the topological constraints of existing SDF-based approaches.