An online system identifies a candidate content item eligible for presentation to a viewing user of the online system, in which the candidate content item includes media (e.g., an image, a video, etc.). The online system identifies one or more media attributes for the media, such as color saturation, tone, brightness, sharpness, contrast, etc. The online system also predicts a value of a performance metric for the candidate content item that indicates a likelihood of user engagement with the candidate content item by the viewing user. For each modification that may be made to a media attribute, the online system predicts a change to the value of the performance metric. Based on the predicted change, the online system determines an optimal set of media attributes associated with a maximum predicted value of the performance metric. The online system modifies the media based on the optimal set of media attributes.