We propose an approach to no-reference image quality evaluation that is consumer-centric and targets real consumer-type images with realistic distortions and realistic quality ranges. We show that the state-of-the-art approaches to no-reference image quality assessment do not perform as well on real consumer-type images, and propose an approach which is simple, efficient, easily interpretable, and achieves high prediction performance on a dataset of images with real non-simulated distortions.