In this paper we show a way to create stable full body motion for a humanoid robot without defining all joint trajectories in advance. The full body motion is split in a task and compensation motion. The task motion can be generated in advance, while the compensation motion is obtained during execution of the task. We show that the compensation motion can be obtained by making use of the linear relation between the joint velocities and the linear momentum of the complete robot. A setpoint for the linear momentum is generated by making use of the error between the current Capture Point and the desired Capture Point location. We have implemented the control algorithm in simulation as well on a real humanoid robot. We were able to create stable full body motion for a bending forward task. It turns out that the control algorithm to create stable full body motion is insensitive for model errors in the internal model of the robot.