The process of generating images with acne is inherently complex, and altering only the acne on an image without distorting the original content is even more challenging. In this context, we present ACNEDIT, a workflow including Generative Adversarial Network (GAN) to modify the severity of acne on facial images derived from user-provided selfies, while preserving realistic skin details and ensuring coherence both among lesions and with the surrounding facial structure. ACNEDIT consists of three key stages: (1) precise analysis of the input image to detect and assess acne characteristics, (2) context-aware overlay of one or more acne lesions based on this analysis, and (3) refinement of this augmented image using a GAN model trained using Acne04 dataset and post-processing techniques to produce a highly realistic result. Our evaluation includes qualitative assessments and a downstream segmentation task (improvement of 7.85% for the IOU and 8.56% for the Dice score), which together demonstrate the high visual quality of the generated images without noticeable artifacts (e.g., deformed lesions or visual distortions). A user study results indicate that acne-synthesized images were often misclassified as unaltered (59.2%), demonstrating the visual realism of our acne generation. We provide generated images that can be used to balance the ACNE04 dataset at https://github.com/AIpourlapeau/ACNEDIT .