Creating human faces that express emotions is something that graphic designers have done for years. When viewing faces that express positive or negative emotions, observers of these faces try to react emotionally due to the fact that certain neural patterns are related to be emphatic, so that he/she tries to produce an emotion linked to the one they are seeing. Using AI, designers are creating human expressions, but the accuracy of the emotion is not necessarily the expected one, because of factors like the intensity of the emotion, the plane of the image, and the prompts arranged to create the intended images. To determine if the same expected emotion produces the same effect in observers, five human images created with Midjourney were tested with Face-Reader software designed by Noldus (v. 9.1). Prompts were altered so as to create different versions of the intended image, modifying instructions concerning specificities of the emotional expressions. Besides, the five images were exposed to 17 observers using the same face-reading software, in order to record their emotional reactions while viewing the images. The results allow us to conclude that human emotional expressions that are intended to be created by means of prompts given to the AI tool Midjourney (v. 5.2), where one of the key-words is the expected emotional expression, cannot be totally accurate, up to a point that different emotions are recognized. In terms of emotions elicited by the images created by Midjourney, observers react in different ways, where it is recognizable how the human faces conceived this way produce emotions of different kinds of emotional valence.