Multi-view image generation methods often suffer from geometric inconsistency and texture discontinuity due to limited 3D scene understanding. This paper presents View-Diff, a novel geometry-aware diffusion-based framework for controllable and high-quality multi-view image generation in aerial scenes. The key innovations comprise two aspects: 1) A Dual-Branch pipeline consists of two parts. The first is a viewpoint transformation branch that precisely transforms known regions. The second is a Spatial Guidance Module branch, which employs scene geometry features to guide the realistic completion of unknown regions and maintain spatial consistency across different viewpoints. 2) an Optimized Perceptual Module incorporates structure-preserving perceptual loss to strengthen feature preservation in known regions, thereby boosting overall coherence. Extensive experiments were conducted on multiple aerial multi-view datasets. View-Diff outperforms state-of-the-art baselines. It achieves notable gains in I2I, LPIPS, and FID on SUES-200 and University-1652. It also shows significant improvements in viewpoint accuracy and visual quality.