For situations like crowd simulation, serious games, and VR-based training, flexible and spontaneous movements are extremely important. Motion models would be the best strategy to adopt, but unfortunately, they are very costly to develop and the results are disappointing. Motion capture is still the most popular way. The ultimate in terms of motion models seems to be data-driven. Motion retargeting and PCA-based models are well used but they still rely strongly to Motion Capture. In this paper, we try to analyze the situation and illustrate it using a few case studies.