We speed up the image generation algorithm of PixelCNN++ by avoiding redundant computation through caching. Naive generation discards computation that can be re-used and performs additional computation that will not be used to generate a particular pixel. Naive generation can take up to 11 minutes to generate 16 32-by-32 images on a Tesla K40 GPU. By re-using previous computation and only performing the minimum amount of computation required, we achieve up to a 183 times speedup over the naive generation algorithm. We have tested our code with Python 3 and TensorFlow 1.0. You may need to make small changes for other versions of Python or TensorFlow.