Abstract: There is significant motivation to provide robot manipulators with improved autonomy using image based visual servo (IBVS) control. In classical approach, an image Jacobian matrix maps image space errors into errors in Cartesian space. Then, a simple proportional control law can be applied guaranteeing local convergence to a desired set point. This control strategy can not deal with delays due to image acquisition and processing and nonlinear constraints such as joint limits and actuator saturation or visibility constraint. In this paper, an IBVS control architecture based on model predictive control (MPC) is presented considering the direct dynamic model of the robot as a Virtual Cartesian Motion Device (VCMD), its joint and torque limits, the camera projection model and the visibility constraint. Considering the VCMD and camera models, the plant model of the visual servo open-loop was derived and, based on it, two predictive control strategies were developed. First, the control approach uses the Generalized Predictive Controller (GPC) to improve accuracy along trajectories in Cartesian space by employing information on the future reference. The second strategy is a predictive control algorithm based on EPSAC (Extended Predictive Self Adaptive Controller) which can take into account the nonlinearities of the plant model. The effectiveness of the predictive control strategies are successfully validated by simulations and with real-time experiments on a 6 DOF industrial robot manipulator.