Abstract: Spot-the-difference is a type of puzzles, where users try to find the different parts of two perceptually similar but actually different images. We propose a semi-automatic system to produce various spot-the-difference puzzle images tagged with their difficulties from a single input image. First, we extract regions to modify from the input by our modified maximal similarity-based region merging algorithm with little user intervention and then apply a variety of image editing techniques for each region to create a modified image. We evaluate the difficulty of a pair of the input and the modified images by considering the saliency and the perceptual difference of the modified region. We provide an empirical model to estimate the time to solve a pair of the images with respect to its difficulty. We show our experimental results and quantitative user research results to evaluate the effectiveness of the proposed method.