A retrat robot or _spoken portrait_ in English (“spoken portrait”) is a graphic reconstruction of the facial features of a person, which is elaborated from the description of someone who knows him. This forensic technique has been used since the 1950s to help identify criminals or missing persons. Although there are currently computer programs that facilitate the same task, often the police, in this case the Mossos d’Esquadra, continue to prefer the skill of a cartoonist who interprets oral testimonies. For an art theory, robot portraiture involves the reverse process of ecphrasis, which consists of translating an image into words. For semiology, it shows both fluency and friction between verbal and visual information. And for the history of imaging technologies, it constitutes a precedent of artificial intelligence “_text to image_” generative devices, where the _prompt_ replaces the instructions of the witness and the algorithm to the cartoonist.