Engineering inverse problems are inherently ill-posed, and various attempts have been made to address them using ML-based surrogate models with heuristic optimization strategies. More recently, generative models have been applied to inverse problems. Among them, denoising diffusion models have attracted particular attention due to their stable sampling behavior and strong expressive power, and these models have been adopted across a wide range of engineering inverse problems. However, most diffusion-based inverse design approaches remain purely data-driven and can generate samples that are not physically consistent [1,2]. As a result, generated designs may appear to satisfy prescribed targets while still requiring additional evaluation to verify whether they actually meet the prescribed target conditions. In this work, we propose a physically consistent inverse design framework based on a physics-informed diffusion model [3] for nonlinear hyperelastic microstructures. The framework aims to reduce the computational cost associated with design verification by jointly modeling microstructural design variables and displacement fields. The training objective integrates a first-principle-based loss derived from weak-form residuals of the governing equations together with a condition loss reflecting target material properties. By embedding physical consistency directly into the generative process, material responses associated with generated designs can be computed without explicitly running nonlinear solvers, while improving the reliability of the evaluated responses. In addition, the proposed framework is further combined with a physics-guided sampling strategy, and its potential for design exploration under data-scarce and out-of-distribution conditions is discussed.