
(a) Copy: the user uses a brush to select a region containing an interesting object or shape, defining the target. (b) Paste: The user positions and pastes the copied object into a single target image. (c) Context: To control generalization, the user selects target regions in several images. (d) The edit is applied to the model, not to a specific image, such that newly generated images will have hats on top of horse heads. (e) The change has been applied to different types of horses and poses.