This paper addresses the challenge of machine unlearning for image-to-image (I2I) generative models, a novel approach to remove sensitive data from machine learning models without retraining. The authors propose a unified framework and an efficient algorithm that effectively removes information from "forget" samples while minimizing performance degradation on "retain" samples. The framework is applicable to various I2I generative models, including diffusion models, VQ-GAN, and MAE. The algorithm is theoretically analyzed and shown to be computationally efficient. Empirical studies on large datasets (ImageNet-1K and Places-365) demonstrate that the proposed method achieves negligible performance degradation on retain sets and effectively removes information from forget sets, even without access to exact retain samples. This work is the first to systematically explore machine unlearning for I2I generative models, providing a comprehensive theoretical and empirical foundation.This paper addresses the challenge of machine unlearning for image-to-image (I2I) generative models, a novel approach to remove sensitive data from machine learning models without retraining. The authors propose a unified framework and an efficient algorithm that effectively removes information from "forget" samples while minimizing performance degradation on "retain" samples. The framework is applicable to various I2I generative models, including diffusion models, VQ-GAN, and MAE. The algorithm is theoretically analyzed and shown to be computationally efficient. Empirical studies on large datasets (ImageNet-1K and Places-365) demonstrate that the proposed method achieves negligible performance degradation on retain sets and effectively removes information from forget sets, even without access to exact retain samples. This work is the first to systematically explore machine unlearning for I2I generative models, providing a comprehensive theoretical and empirical foundation.