This paper addresses the problem of obtaining real-time parameterized simulations of human motion thanks to motion capture data. Contrary to motion graphs that are based on large database of motions, our method relies on per-frame adaptations of motion clips that are selected by a user. For the locomotion example, the user selects a motion that conveys a specific style and drives the character in a complex environment in real-time with only the information of stances that can be computed automatically. The framework presented in this paper uses then sequentially several features such as morphological adaptation, kinematic constraints solving and dynamic correction of the motion. It automatically adapts the required motions to the size of the synthetic figure and to its environment (such as complex grounds and additional external forces). All these features are organized in layers in order to be easily combined together. It also simplifies the design of controllers such as those enabling displacements or grasping. Indeed, since all the features are embedded and organized in layers, creating a controller simply means defining the constraints (kinematic, dynamic...) and the framework ensures animating hundreds of humanoids with different morphologies in real-time (perframe solver). It is particularly suitable for interactive applications such as video games and virtual reality where a user interacts in an unpredictable way.