Effects of anthropomorphism or zoomorphism in social robotics motivate two opposing tendencies in the philosophy and ethics of robots: a ‘rational’ tendency that discourages excessive anthropomorphism because it is based on an illusion and a ‘visionary’ tendency that promotes the relational reality of human-robot interaction. I argue for two claims: First, the opposition between these tendencies cannot be resolved and leads to a kind of technological antinomy. Second, we can deal with this antinomy by way of an analogy between our treatment of robots as social interactors and the perception of objects in pictures according to a phenomenological theory of image perception. Following this analogy, human- or animal-likeness in social robots is interpreted neither as a psychological illusion, nor as a relational reality. Instead, robots belong to a special ontological category shaped by perception and interaction, similar to objects in images.