This paper investigates the mechanisms and timing of how negative prompts affect image generation in diffusion models. The study reveals two key behaviors: delayed effect, where negative prompts influence the generation process after positive prompts have already rendered content, and deletion through neutralization, where negative prompts cancel out positive prompts in the latent space. These findings have significant implications for real-world applications, such as object inpainting, where negative prompts can be used to remove unwanted elements with minimal background alteration. The research explores the dynamics of diffusion steps and identifies the critical step at which negative prompts begin to influence the generation process. It also uncovers a phenomenon called "reverse activation," where introducing a negative prompt early in the diffusion process paradoxically leads to the generation of the specified object. This is explained by the "inducing effect" and "momentum effect" in diffusion dynamics, where negative prompts can induce positive noise in specific directions and maintain a consistent direction in the diffusion process. The study proposes a novel approach for controllable image inpainting, where negative prompts are applied after the critical step in the reverse-diffusion process. This method preserves the original background while effectively removing undesired elements. The approach is training-free and does not require modifications to the diffusion model during inference. The paper also introduces a new metric to quantify the strength of negative prompts and conducts extensive experiments to validate the effectiveness of the proposed method. Results show that the method achieves high removal success rates and maintains strong similarity to the original images. The findings contribute to a deeper understanding of negative prompts in diffusion models and offer practical solutions for controllable image inpainting tasks.This paper investigates the mechanisms and timing of how negative prompts affect image generation in diffusion models. The study reveals two key behaviors: delayed effect, where negative prompts influence the generation process after positive prompts have already rendered content, and deletion through neutralization, where negative prompts cancel out positive prompts in the latent space. These findings have significant implications for real-world applications, such as object inpainting, where negative prompts can be used to remove unwanted elements with minimal background alteration. The research explores the dynamics of diffusion steps and identifies the critical step at which negative prompts begin to influence the generation process. It also uncovers a phenomenon called "reverse activation," where introducing a negative prompt early in the diffusion process paradoxically leads to the generation of the specified object. This is explained by the "inducing effect" and "momentum effect" in diffusion dynamics, where negative prompts can induce positive noise in specific directions and maintain a consistent direction in the diffusion process. The study proposes a novel approach for controllable image inpainting, where negative prompts are applied after the critical step in the reverse-diffusion process. This method preserves the original background while effectively removing undesired elements. The approach is training-free and does not require modifications to the diffusion model during inference. The paper also introduces a new metric to quantify the strength of negative prompts and conducts extensive experiments to validate the effectiveness of the proposed method. Results show that the method achieves high removal success rates and maintains strong similarity to the original images. The findings contribute to a deeper understanding of negative prompts in diffusion models and offer practical solutions for controllable image inpainting tasks.