Vision-based parking slot occupancy detection is a key component of intelligent transportation systems, but its performance degrades under adverse weather conditions such as rain, fog, and snow. Public datasets such as PKLot and CNRPark-EXT contain limited variability in severe weather, making robust generalization difficult. This paper introduces a weather-aware data augmentation pipeline based on Qwen-Image-Edit, a diffusion-based vision-language image editing model. The method generates synthetic snow, fog, and rain variants from a subset of overcast images using natural-language instructions, while preserving parking geometry and slot-level annotations; however, careful prompt design is required to ensure structural consistency during image editing. A patch-level ResNet-50 classifier is retrained on the augmented dataset and evaluated in a cross-dataset setting from CNRPark-EXT to PKLot. Experimental results show improvements of up to 4.0 percentage points in accuracy and up to 10.6 percentage points in precision, with consistent gains across all weather subsets in F1-score and accuracy.