With the rapid advancement of photographic technology, the demand for improving image quality and visual appeal in challenging conditions is growing. This paper investigates an innovative approach to image enhancement in low-light scenes by utilizing a deep learning framework that combines Transformer and U-Net (UNet) architectures. Specifically, a hierarchical encoder-decoder UNet Transformer (Uformer) model integrates the strengths of UNet's structural design with the multi-scale self-attention mechanisms of Transformers to capture intricate dependencies across feature maps at various scales, enabling superior recovery of image details and enhancement at multiple resolutions. This study further introduces a novel loss function combination—incorporating edge loss, Peak Signal-to-Noise Ratio (PSNR) loss, Structural Similarity Index Measure (SSIM) loss, and Charbonnier loss—to optimize the enhancement process more comprehensively. The combined Transformer-UNet approach, together with the multi-loss strategy, offers a significant improvement in detail preservation and overall image quality under low-light conditions. Through extensive ablation and comparative experiments, this paper demonstrates the effectiveness of this hybrid architecture and the proposed loss functions, providing valuable insights and methodological advancements for future research in image enhancement.