<p>Quantum Generative Adversarial Networks (QGANs) represent a viable approach for data generation on noisy intermediate-scale quantum devices. Nevertheless, established QGAN architectures face considerable challenges in processing high-dimensional structured data, such as images, owing to the incompatibility between static quantum generators and the pronounced spatial interdependencies characteristic of image distributions. Given the constraints on quantum resources, such limitations frequently lead to training instability and diminished structural integrity. In response, this study introduces a recurrent quantum generative adversarial network that reconceptualizes image generation as a sequential procedure. Drawing upon principles from quantum long short-term memory (QLSTM), the generator employs a dynamic quantum hidden state throughout patch-wise generation phases, thereby facilitating the transmission of structural details without expanding qubit requirements or circuit complexity. Evaluations conducted on the MNIST and Fashion-MNIST datasets reveal that this framework surpasses selected benchmark QGAN models in visual fidelity and objective measures, reflecting enhanced structural uniformity and operational reliability under resource restrictions. The inherently large-scale nature of parallel quantum circuit simulation and distributed parameter-shift gradient estimation in this hybrid quantum-classical pipeline necessitates high-performance computing infrastructure, positioning this work as both a methodological advance in quantum generative modeling and a representative use case for modern supercomputing facilities.</p>