Fine-Grained Sketch-Based Image Retrieval (FG-SBIR) is challenging due to the domain gap between sketches and photos, large intra-class variations, and subtle inter-class differences. We introduce Hybrid Attention-Guided FG-SBIR (HAG-FG-SBIR), a framework that integrates multi-scale feature fusion, hybrid attention, and prototype-based alignment. Our design combines self-attention for intra-modal enhancement, cross-modal bi-directional attention for explicit sketch–photo alignment, and channel attention for emphasizing discriminative features. To further improve fine-grained discrimination, we propose an instance-level prototype contrastive alignment that pulls sketches and photos toward shared prototypes. Training is guided by a composite loss unifying triplet ranking, prototype contrastive, and attention regularization. Experiments on four benchmarks demonstrate that HAG-FG-SBIR achieves state-of-the-art accuracy with lower computational cost, while attention maps reveal interpretable focus on semantically meaningful regions.