Deep Neural Networks (DNNs) are increasingly integrated into safety-critical systems, such as autonomous vehicles and planetary exploration robots. However, ensuring their reliability remains a challenge due to the limitations of traditional testing methods, which rely on low-fidelity simulations. This paper introduces DESIGNATE, a novel framework that combines search-based testing with Generative Adversarial Networks (GANs) to generate realistic failure-inducing test cases. DESIGNATE employs meta-heuristic search to explore input space efficiently, GANs to enhance simulation fidelity, and a multi-objective genetic algorithm to optimize both accuracy and diversity in test inputs. We evaluate DESIGNATE on two case studies: urban road scene segmentation using DeepLabV3+ and Martian object detection using the AI4MARS dataset. Empirical results show that DESIGNATE outperforms state-of-the-art approaches, identifying diverse failure cases and achieving improvements in DNN accuracy after retraining. Our findings highlight the necessity of integrating GANs with simulator-based testing to enhance DNN robustness in real-world applications.