Issue with Image Modification in ChatGPT-4

I am writing to report an issue I encountered while using ChatGPT-4 for generating images. When I request a specific alteration to an existing image, such as changing the position of a character or modifying a small detail, the tool often generates an entirely new image that does not retain the elements of the original. This has been frustrating as I am looking for minor modifications rather than a complete recreation of the image.

For example, I recently requested that an image of a character have only the arm position changed. However, the resulting image completely changed the background, the character’s style, and other details that were not supposed to be altered.

I would like to know if there is any way to improve control over specific modifications in generated images, or if there are steps I can follow to ensure that only the requested details are altered. I appreciate your attention to this issue and hope you can help improve this functionality.