We propose a novel approach for a dense texture-based visualization of vector fields on curved surfaces. Our texture advection mechanism relies on a Lagrangian particle tracing that is simultaneously computed in the physical space of the object and in the device space of the image plane. This approach retains the benefits of previous image-space techniques, such as output sensitivity, independence from surface parameterization or mesh connectivity, and support for dynamic surfaces. At the same time, frame-to-frame coherence is achieved even when the camera position is changed, and potential inflow issues at silhouette lines are overcome. Noise input for texture advection is modeled as a solid 3D texture and constant spatial noise frequency on the image plane is achieved in a memory-efficient way by appropriately scaling the noise in physical space. For the final rendering, we propose color schemes to effectively combine the visualization of surface shape and flow. Hybrid physical/devi...