ZipAR: Parallel Autoregressive Image Generation through Spatial Locality

In this paper, we propose ZipAR, a training-free, plug-and-play parallel decoding framework for accelerating autoregressive (AR) visual generation. The motivation stems from the observation that images exhibit local structures, and spatially distant regions tend to have minimal interdependence. Given a partially decoded set of visual tokens, in addition to the original next-token prediction scheme in the row dimension, the tokens corresponding to spatially adjacent regions in the column dimension can be decoded in parallel. To ensure alignment with the contextual requirements of each token, we employ an adaptive local window assignment scheme with rejection sampling analogous to speculative decoding. By decoding multiple tokens in a single forward pass, the number of forward passes required to generate an image is significantly reduced, resulting in a substantial improvement in generation efficiency. Experiments demonstrate that ZipAR can reduce the number of model forward passes by up to 91% on the Emu3-Gen model without requiring any additional retraining.

Generating images with AI models is slow because these models typically create one small piece of the image at a time, in sequence. This step-by-step process requires many computations, making it inefficient—especially for high-resolution images. We developed ZipAR, a method that speeds up image generation by predicting multiple pieces of the image simultaneously, without retraining the model. Our key insight is that distant parts of an image (like the top and bottom) often don’t depend on each other, so they can be generated in parallel. To ensure these parallel predictions stay coherent, we use an adaptive "local window" technique—similar to how a painter might sketch rough outlines of separate sections. ZipAR reduces the number of computations needed by up to 91% in some cases, drastically speeding up image generation while maintaining quality. This makes AI tools for art, design, and media production faster and more practical to use.