<p>Sketch-to-image generation is a crucial task in image synthesis, yet existing methods often suffer from category confusion and inconsistent textures due to the abstraction of sketches. To address these limitations, we propose a novel edge-enhanced sketch-to-image generation framework with three key components. First, we introduce an edge-driven strategy that transforms abstract sketches into intermediate edge representations, providing structural guidance and narrowing the domain gap. Second, we integrate spatial, channel, and multi-head attention mechanisms into the real image generator to enhance fine-grained detail while preserving global structure. Third, we design an AttenAdaInResnetBlock that incorporates label information into adaptive instance normalization, ensuring semantic consistency and coherent textures across categories. Extensive experiments on Scribble and QMUL-Sketch benchmarks demonstrate that our method achieves higher category accuracy and improved realism compared with state-of-the-art approaches, validating the effectiveness of our edge-driven design.</p>