EmoNet, a new convolutional neural network, can accurately decode images into eleven distinct emotional categories. Training the AI on over 25,000 images, researchers demonstrate image content is sufficient to predict the category and valence of human emotions.