How can we identify and track violent tactics within protest events? Protests are usually classified as peaceful or violent based on the most salient form of participants’ behaviour. However, we do not know much about what else happens within protest events that may eventually lead to the appearance of violent tactics. We present a spectrum of violent tactics that participants can use to show dissent against the government. We argue that while some protests display a consistent preference for peaceful or violent resistance from the outset, others show diversity in their behaviour, adding uncertainty to their overall outcome. Harnessing a new image-to-data methodology, we analyze 15,567 protest images across 10 countries between 2016 and 2021. In doing so, we provide two contributions to the field of organized dissent. First, we explain how within protest violence can be identified using the composition of particular objects. We do so by leveraging the information of over 8.1 million objects detected on the protest images using the Large Vocabulary Instance Segmentation (LVIS) dataset. Second, we extract and classify image-specific features on a diverse pool of participant behaviour over our entire sample, allowing us to recognize violent events using images. Overall, this new data collection effort and analysis help us gain insights into the inner workings of protest events.