AccDiffusion is a novel method for generating higher-resolution images without training, addressing the issue of object repetition in patch-wise image generation. The method decouples the vanilla image-content-aware prompt into a set of patch-content-aware prompts, ensuring each patch receives a precise description. Additionally, AccDiffusion introduces dilated sampling with window interaction to enhance global consistency in higher-resolution image generation. Experimental results demonstrate that AccDiffusion effectively reduces object repetition and achieves state-of-the-art performance in training-free image generation extrapolation. The method is evaluated using various metrics and compared with existing techniques, showing superior results in both quantitative and qualitative assessments.AccDiffusion is a novel method for generating higher-resolution images without training, addressing the issue of object repetition in patch-wise image generation. The method decouples the vanilla image-content-aware prompt into a set of patch-content-aware prompts, ensuring each patch receives a precise description. Additionally, AccDiffusion introduces dilated sampling with window interaction to enhance global consistency in higher-resolution image generation. Experimental results demonstrate that AccDiffusion effectively reduces object repetition and achieves state-of-the-art performance in training-free image generation extrapolation. The method is evaluated using various metrics and compared with existing techniques, showing superior results in both quantitative and qualitative assessments.