Social group categorization has been mainly studied in relation to ownership manipulations involving highly-salient multisensory cues. Here, we propose a novel paradigm that can implicitly activate the embodiment process in the presence of group affiliation information, whilst participants complete a task irrelevant to social categorization. Ethnically White participants watched videos of White- and Black-skinned models writing a proverb. The writing was interrupted 7, 4 or 1 s before completion. Participants were tasked with estimating the residual duration following interruption. A video showing only hand kinematic traces acted as a control condition. Residual duration estimates for out-group and control videos were significantly lower than those for in-group videos only for the longest duration. Moreover, stronger implicit racial bias was negatively correlated to estimates of residual duration for out-group videos. The underestimation bias for the out-group condition might be mediated by implicit embodiment, affective and attentional processes, and finalized to a rapid out-group categorization.