Diffusion models are powerful tools for generating high-quality images but pose risks such as the unintentional generation of inappropriate content. This paper introduces EraseDiff, an algorithm designed to preserve the utility of diffusion models while removing unwanted information. EraseDiff formulates the task as a constrained optimization problem using the value function, achieving a balance between preservation and erasure. Extensive experiments demonstrate that EraseDiff effectively preserves model utility and efficiency, outperforming state-of-the-art methods in various scenarios, including class-wise and concept-wise forgetting. The algorithm is efficient, with a speed increase of up to 11 times compared to existing methods, and maintains a good trade-off between erasing and preserving model utility. However, the scrubbed models may still retain some characteristics of the forgotten data, and future work could address issues like fairness and advanced privacy-preserving techniques.Diffusion models are powerful tools for generating high-quality images but pose risks such as the unintentional generation of inappropriate content. This paper introduces EraseDiff, an algorithm designed to preserve the utility of diffusion models while removing unwanted information. EraseDiff formulates the task as a constrained optimization problem using the value function, achieving a balance between preservation and erasure. Extensive experiments demonstrate that EraseDiff effectively preserves model utility and efficiency, outperforming state-of-the-art methods in various scenarios, including class-wise and concept-wise forgetting. The algorithm is efficient, with a speed increase of up to 11 times compared to existing methods, and maintains a good trade-off between erasing and preserving model utility. However, the scrubbed models may still retain some characteristics of the forgotten data, and future work could address issues like fairness and advanced privacy-preserving techniques.