A growing number of users in Web 2.0 based social network sites and photo sharing portals upload millions of images per day. In many cases, this leads to serious privacy threats. The images reveal not only the personal relationships and attitudes of the uploader, but of other persons displayed in the images as well. In this paper, we propose the PRIMO system architecture for privacy-aware image sharing. Our approach is based on semantic annotations, face recognition and user-defined privacy rules. PRIMO connects to many social network sites and photo sharing portals via the OpenSocial API and proprietary interfaces.