Visual Anagrams is a simple, zero-shot method for generating multi-view visual illusions. We demonstrate both theoretical and practical evidence that our method supports a wide range of transformations, including rotation, flipping, color inversion, tilting, puzzle rearrangement, and random shuffling. Our method uses a pre-trained diffusion model to estimate the noise in different views or transformations of an image and aligns and averages it. This averaged noise estimate is then used to perform the diffusion steps. Using Visual Anagrams, you can create a variety of multi-view visual illusions.