Directly resolving in real-space multiple atomic motions using ultrafast x-ray scattering is generally limited by the finite detector range. As a result, signal interpretation mostly relies on modeling and simulations of specific excitation pathways. Here, we demonstrate an approach to resolve ultrafast diffuse x-ray scattering signals in real space and recover multiple atomic motions de-novo. We introduce a scattering basis representation that is composed of the measurement parameters and constraints, and the subsequent inversion analysis. We then leverage signal priors, such as smoothness and sparsity to deconvolve and super-resolve the spatially transformed signals using convex optimization. We validate the approach on simulated data with detection limits similar to X-ray free-electron laser experiments and discuss the resolution limits and noise dependence on the accuracy of the recovery.