<p>Quantum generative adversarial networks, a key branch of quantum machine learning, show promise in image generation but face constraints from Noisy Intermediate-Scale Quantum hardware, particularly for high-resolution tasks. To overcome this, we propose HQWGAN, a hybrid quantum–classical architecture that leverages high-performance computing for scalable quantum circuit simulations, parallel sub-generator processing, and real-time noise control, enabling efficient training on large datasets. The model uses principal component analysis for dimensionality reduction, introduces a Reverse-S-shaped allocation strategy to address high-dimensional feature learning limitations, and incorporates an Exponential Smoothed Noise Control mechanism to enhance diversity and stability by adaptively adjusting quantum circuit perturbations, mitigating mode collapse. The generator employs highly expressive and entangled quantum circuits to capture complex patterns under resource constraints. Experiments on MNIST and Fashion-MNIST datasets show HQWGAN outperforms existing methods in image quality and robustness, validating its mechanisms.</p>