SCAFFOLD is a simple and general visual prompting method designed to enhance vision-language coordination in Large Multi-Modal Models (LMMs). The method overlays a dot matrix on images, with each dot labeled with multi-dimensional Cartesian coordinates. These coordinates are also included in the textual prompt, helping LMMs align visual and textual information. SCAFFOLD has been tested on various challenging vision-language tasks, including spatial reasoning, compositional reasoning, fine-grained visual understanding, and hallucination detection. The results show that SCAFFOLD outperforms existing methods like GPT-4V with textual CoT prompting. The method is also effective in combination with other techniques such as active perception and Chain-of-Thought prompting. SCAFFOLD's effectiveness is further supported by ablation studies, which show that the method's performance is robust to variations in matrix size and coordinate format. The method is also adaptable to different scenarios, with the potential for further improvements through dynamic adjustments. The study highlights the importance of visual prompting in enhancing LMM capabilities and suggests that future research should focus on improving visual localization and grounding in complex environments.SCAFFOLD is a simple and general visual prompting method designed to enhance vision-language coordination in Large Multi-Modal Models (LMMs). The method overlays a dot matrix on images, with each dot labeled with multi-dimensional Cartesian coordinates. These coordinates are also included in the textual prompt, helping LMMs align visual and textual information. SCAFFOLD has been tested on various challenging vision-language tasks, including spatial reasoning, compositional reasoning, fine-grained visual understanding, and hallucination detection. The results show that SCAFFOLD outperforms existing methods like GPT-4V with textual CoT prompting. The method is also effective in combination with other techniques such as active perception and Chain-of-Thought prompting. SCAFFOLD's effectiveness is further supported by ablation studies, which show that the method's performance is robust to variations in matrix size and coordinate format. The method is also adaptable to different scenarios, with the potential for further improvements through dynamic adjustments. The study highlights the importance of visual prompting in enhancing LMM capabilities and suggests that future research should focus on improving visual localization and grounding in complex environments.