This paper introduces a novel method for text-driven editing of natural images, leveraging both CLIP and DDPM models. The method allows for region-based edits using natural language descriptions and ROI masks, ensuring seamless integration of the edited region with the rest of the image. By blending noised versions of the input image with the text-guided diffusion latent at various noise levels, the method achieves realistic and coherent results. The authors also introduce a technique to mitigate adversarial effects by augmenting the diffusion process. The method outperforms existing baselines in terms of realism, background preservation, and text matching. The paper demonstrates various applications, including object addition, removal, replacement, background replacement, and image extrapolation. The main contributions include the first solution for general-purpose region-based image editing using natural language guidance, background preservation, and the effectiveness of extending augmentations. The limitations discussed include inference time and the need for further improvements in ranking methods and addressing biases in CLIP. The societal impact of the method is also explored, highlighting both potential benefits and risks.This paper introduces a novel method for text-driven editing of natural images, leveraging both CLIP and DDPM models. The method allows for region-based edits using natural language descriptions and ROI masks, ensuring seamless integration of the edited region with the rest of the image. By blending noised versions of the input image with the text-guided diffusion latent at various noise levels, the method achieves realistic and coherent results. The authors also introduce a technique to mitigate adversarial effects by augmenting the diffusion process. The method outperforms existing baselines in terms of realism, background preservation, and text matching. The paper demonstrates various applications, including object addition, removal, replacement, background replacement, and image extrapolation. The main contributions include the first solution for general-purpose region-based image editing using natural language guidance, background preservation, and the effectiveness of extending augmentations. The limitations discussed include inference time and the need for further improvements in ranking methods and addressing biases in CLIP. The societal impact of the method is also explored, highlighting both potential benefits and risks.