Visual working memory (VWM) is one of the most crucial parts of the human cognitive system. Research focuses on the apparent limits in the capacity of this system and the reasons for them. But so far only a few formal models exist that can account for the temporal dynamics of the amount of information stored in VWM. We propose a combination of the well established theory of visual attention (TVA) with a dynamic memory model, resulting in an iterative, probabilistic framework for VWM. The model includes a consolidation as well as a decay mechanism and employs the strength concept to quantify the availability of a certain memory trace. We evaluate the model on available change detection data.