We are constantly surrounded by photographs: on social media, in advertisements, and many other aspects of life. Such images may be perceived as beautiful because of their very content (e.g., a spectacular rainbow) or their excellent quality (e.g., a well-executed photograph of an everyday object). However, many studies on visual aesthetics conflate these factors, making it difficult to determine to what extent the quality of an image influences its perceived beauty, and whether participants are able to judge an image’s beauty independently of its quality. Focusing on scene perception, we algorithmically altered the image quality of a set of natural scene photographs. We created three versions of each image: images were ‘beautified’ by enhancing their saturation and upsampling them using a super-resolution deep neural network, ‘uglified’ by adding noise and reducing color fidelity, or left virtually unedited. Participants then rated these images in two behavioral studies. In the first study, participants rated each scene for its beauty. Here, uglified scenes were reliably rated as less beautiful than both beautified and unedited scenes. By contrast, beautification did not reliably increase beauty ratings for all images. In the second study, participants rated scene beauty and image quality at the same time. This dual task reduced the influence of image quality on beauty ratings, suggesting that participants can judge scene beauty and image quality relatively independently of each other. A simple concurrent judgment of image quality may thus be useful for experimentally disentangling beauty and image quality.