Aspects of the present disclosure involve a system for performing real-time in-painting using machine learning techniques. The system receives a video that includes a depiction of a real-world object in a real-world environment. The system accesses a segmentation associated with the real-world object and removes a depiction of the real-world object from a region of a first frame of the video. The system processes, by a machine learning model, the first frame and one or more previous frames of the video that precede the first frame to generate a new frame in which portions of the first frame have been blended into the region from which the depiction of the real-world object has been removed.