ObjectDrop is a supervised method aimed at achieving photorealistic object removal and insertion. It utilizes a counting fact dataset and self-supervised learning techniques. Its main functions include removing objects from images, along with their impact on the scene (such as occlusion, shadows, and reflections), and inserting objects into images in an extremely realistic manner. It achieves object removal by fine-tuning a diffusion model on a small, specialized captured dataset. For object insertion, it employs a self-supervised approach using the removal model to synthesize a large-scale counting fact dataset, which is then used to train and further fine-tune the insertion model on real datasets, resulting in high-quality insertion results. Compared to previous methods, ObjectDrop demonstrates significant advancements in the realism of both object removal and insertion.