There currently exist a number of excellent techniques to capture, register, and blend digital panoramas. However, the problem is treated as an automated batch process, which can take several minutes to produce a panorama. Unfortunately, many of the algorithms involved are prone to errors and/or artifacts and may require meticulous tuning to achieve high quality results. This not only requires expert knowledge, but given the complex techniques involved can also result in a tedious and time consuming trial-and-error process. Each update may influence previous corrections and take minutes to be computed. Previews are typically not available or, at best, are provided as unintuitive outputs from each batch process. As a result, the existing workflow to create a panorama can be exasperating and exploring different aesthetic choices, such as image selection, etc., is too time consuming to be feasible. In this paper, we move from the traditional inflexible and sequential batch creation to a more versatile, interactive approach. We introduce novel techniques to enable a user-driven panorama workflow that leverages quick, meaningful previews and stream processing to provide the first end-to-end, interactive creation pipeline. This new workflow provides users with online control of the acquisition, registration, and composition without any constraints on input or use of specialized hardware and allows for the first time unconstrained, in-the-field panorama creation on commodity hardware. In particular, our approach is based on: (i) a new registration acceleration scheme to provide instant feedback independent of the number or structure of images in the panorama; (ii) a new mesh data structure to support arbitrary image arrangements; and (iii) a new scheme to provide previews of and progressively stream seam calculations.