An online system presents a user with a content item including an image and text. The online system overlays the text on the image if it predicts doing so will not render the text unreadable. To predict whether the overlay will render the text unreadable, the online system computes values describing various visual characteristics of a segment of the image on which the text may be overlaid and compares the values to corresponding threshold values. Based on the comparison, the online system predicts a readability of the text if overlaid on the segment. If the predicted readability is at least a threshold readability, the online system overlays the text on the segment of the image and presents the content item to the user with the text overlaid on the image. Otherwise, the online system presents the content item to the user with the text not overlaid on the image.