The control of formations of multi-agent systems has profound applications on today's technological scene, ranging from satellite constellations, collaborative load transportation, cooperative surveillance and distributed aperture imaging systems. Often, these applications are needed in environments where localization is challenging or inexistent, such as indoor and underground environment, or extra-planetary scenarios (such as Mars or the Moon). In this work, a novel formation control scheme using image features correspondences from widespread on-board cameras is proposed. The control scheme relies on 5 feature correspondences between two or three agents, and one range measurement between any agent and the formation leader. Then, optimal control inputs generated with a Nonlinear Model Predictive Control-based control law drive the agents, with asymptotic stability guarantees, towards the desired formation setting. The framework is tested both in simulation and on mobile platforms in a laboratory environment, with multiple camera types - from omnidirectional, to fish-eye and perspective imaging systems.