Generating "Spot the Difference" Puzzles with AI (kamens.com)

A developer shares his journey into using AI to generate "Spot the Difference" puzzles, highlighting the significant challenges in ensuring the images produced contain a reliable number of subtle differences. Initial attempts using standard prompts yielded inconsistent results, prompting the need for a more structured approach that combines advanced AI techniques with custom coding. This method enhances the reliability of generated puzzles, which is crucial for their use in children's activity books. The innovative solution employs segmentation via SAM2 to divide the image into manageable objects before applying targeted inpainting to create visible differences. By tuning various parameters, the developer can filter the segments for size and position, ultimately ensuring the differences are challenging yet not impossible for young users to spot. Crucially, perceptual difference scoring techniques, such as LPIPS and color difference algorithms, help ensure the changes are distinct to the human eye. This approach not only bolsters the quality of children's puzzles but also exemplifies a broader trend in the AI/ML community toward integrating precision controls with generative models to meet specific application needs.