In this paper, we propose ZipAR, a training-free, plug-and-play parallel decoding framework for accelerating auto-regressive (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, enabling the ``next-set prediction'' paradigm. By decoding multiple tokens simultaneously 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. Code is available here: https://github.com/ThisisBillhe/ZipAR.
@article{arxiv.2412.04062,
title = {ZipAR: Parallel Auto-regressive Image Generation through Spatial Locality},
author = {Yefei He and Feng Chen and Yuanyu He and Shaoxuan He and Hong Zhou and Kaipeng Zhang and Bohan Zhuang},
journal= {arXiv preprint arXiv:2412.04062},
year = {2025}
}