Cutting out and object and estimate its transparency mask is a key task in many applications. We take on the work on closed-form matting by Levin et al., that is used at the core of many matting techniques, and propose an alternative formulation that offers more flexible controls over the matting priors. We also show that this new approach is efficient at upscaling transparency maps from coarse estimates.