<p>The emergence of generative adversarial networks (GANs) has revolutionized text-to-image synthesis. However, conventional single-level GAN architectures, exemplified by DF-GAN, often struggle to generate fine-grained details, primarily due to insufficient utilization of word-level textual information. To overcome this limitation, we propose the dynamic local affine generative adversarial network (DLA-GAN), which introduces a novel dynamic local affine transformation mechanism. This mechanism effectively incorporates word-level text encoding and computes multiple affine transformations in parallel, while dynamically weighting their contributions through a multilayer perceptron. By integrating this approach, DLA-GAN more accurately captures semantic textual attributes and generates visually coherent images with enhanced fine-grained details. Comprehensive experimental evaluations on multiple benchmark datasets demonstrate that DLA-GAN consistently outperforms existing methods in terms of fine-grained image quality and semantic fidelity to textual descriptions, while maintaining comparable computational efficiency. Our proposed framework advances the state of the art in text-to-image synthesis by effectively leveraging word-level textual information without introducing significant additional computational complexity. The code is available on <a href="https://github.com/TRICKticky/DLA-GAN">https://github.com/TRICKticky/DLA-GAN</a>.</p>