Face detection is a challenging problem having a wide range of security and surveillance-based applications. Skin color can be an important differentiator and used to augment the performance of automatic face detection. However, obtaining skin color consistency across illumination, different camera settings, and diverse ethnicity is a challenging task. Static skin color models that rely on image preprocessing are able to bring only limited consistency. Their performance in terms of accuracy and computation time degrades severely in real-world videos. In this paper, we study the dynamics of different color models on a database of five videos containing more than 93,000 manually annotated face images. Further, we propose an adaptive skin color model to reduce the false accept cases of Adaboost face detector. Since the face color distribution model is regularly updated using previous Adaboost responses, we find the system to be more effective to real-world environmental covariates. Importantly, the adaptive nature of the skin classifier does not significantly affect the computation time. © Springer India 2016.