Token-Shuffle: Towards High-Resolution Image Generation with Autoregressive Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Ma, Xu, Sun, Peize, Ma, Haoyu, Tang, Hao, Ma, Chih-Yao, Wang, Jialiang, Li, Kunpeng, Dai, Xiaoliang, Shi, Yujun, Ju, Xuan, Hu, Yushi, Sanakoyeu, Artsiom, Juefei-Xu, Felix, Hou, Ji, Tian, Junjiao, Xu, Tao, Hou, Tingbo, Liu, Yen-Cheng, He, Zecheng, He, Zijian, Feiszli, Matt, Zhang, Peizhao, Vajda, Peter, Tsai, Sam, Fu, Yun
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908339780190208
author Ma, Xu
Sun, Peize
Ma, Haoyu
Tang, Hao
Ma, Chih-Yao
Wang, Jialiang
Li, Kunpeng
Dai, Xiaoliang
Shi, Yujun
Ju, Xuan
Hu, Yushi
Sanakoyeu, Artsiom
Juefei-Xu, Felix
Hou, Ji
Tian, Junjiao
Xu, Tao
Hou, Tingbo
Liu, Yen-Cheng
He, Zecheng
He, Zijian
Feiszli, Matt
Zhang, Peizhao
Vajda, Peter
Tsai, Sam
Fu, Yun
author_facet Ma, Xu
Sun, Peize
Ma, Haoyu
Tang, Hao
Ma, Chih-Yao
Wang, Jialiang
Li, Kunpeng
Dai, Xiaoliang
Shi, Yujun
Ju, Xuan
Hu, Yushi
Sanakoyeu, Artsiom
Juefei-Xu, Felix
Hou, Ji
Tian, Junjiao
Xu, Tao
Hou, Tingbo
Liu, Yen-Cheng
He, Zecheng
He, Zijian
Feiszli, Matt
Zhang, Peizhao
Vajda, Peter
Tsai, Sam
Fu, Yun
contents Autoregressive (AR) models, long dominant in language generation, are increasingly applied to image synthesis but are often considered less competitive than Diffusion-based models. A primary limitation is the substantial number of image tokens required for AR models, which constrains both training and inference efficiency, as well as image resolution. To address this, we present Token-Shuffle, a novel yet simple method that reduces the number of image tokens in Transformer. Our key insight is the dimensional redundancy of visual vocabularies in Multimodal Large Language Models (MLLMs), where low-dimensional visual codes from visual encoder are directly mapped to high-dimensional language vocabularies. Leveraging this, we consider two key operations: token-shuffle, which merges spatially local tokens along channel dimension to decrease the input token number, and token-unshuffle, which untangles the inferred tokens after Transformer blocks to restore the spatial arrangement for output. Jointly training with textual prompts, our strategy requires no additional pretrained text-encoder and enables MLLMs to support extremely high-resolution image synthesis in a unified next-token prediction way while maintaining efficient training and inference. For the first time, we push the boundary of AR text-to-image generation to a resolution of 2048x2048 with gratifying generation performance. In GenAI-benchmark, our 2.7B model achieves 0.77 overall score on hard prompts, outperforming AR models LlamaGen by 0.18 and diffusion models LDM by 0.15. Exhaustive large-scale human evaluations also demonstrate our prominent image generation ability in terms of text-alignment, visual flaw, and visual appearance. We hope that Token-Shuffle can serve as a foundational design for efficient high-resolution image generation within MLLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2504_17789
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Token-Shuffle: Towards High-Resolution Image Generation with Autoregressive Models
Ma, Xu
Sun, Peize
Ma, Haoyu
Tang, Hao
Ma, Chih-Yao
Wang, Jialiang
Li, Kunpeng
Dai, Xiaoliang
Shi, Yujun
Ju, Xuan
Hu, Yushi
Sanakoyeu, Artsiom
Juefei-Xu, Felix
Hou, Ji
Tian, Junjiao
Xu, Tao
Hou, Tingbo
Liu, Yen-Cheng
He, Zecheng
He, Zijian
Feiszli, Matt
Zhang, Peizhao
Vajda, Peter
Tsai, Sam
Fu, Yun
Computer Vision and Pattern Recognition
Autoregressive (AR) models, long dominant in language generation, are increasingly applied to image synthesis but are often considered less competitive than Diffusion-based models. A primary limitation is the substantial number of image tokens required for AR models, which constrains both training and inference efficiency, as well as image resolution. To address this, we present Token-Shuffle, a novel yet simple method that reduces the number of image tokens in Transformer. Our key insight is the dimensional redundancy of visual vocabularies in Multimodal Large Language Models (MLLMs), where low-dimensional visual codes from visual encoder are directly mapped to high-dimensional language vocabularies. Leveraging this, we consider two key operations: token-shuffle, which merges spatially local tokens along channel dimension to decrease the input token number, and token-unshuffle, which untangles the inferred tokens after Transformer blocks to restore the spatial arrangement for output. Jointly training with textual prompts, our strategy requires no additional pretrained text-encoder and enables MLLMs to support extremely high-resolution image synthesis in a unified next-token prediction way while maintaining efficient training and inference. For the first time, we push the boundary of AR text-to-image generation to a resolution of 2048x2048 with gratifying generation performance. In GenAI-benchmark, our 2.7B model achieves 0.77 overall score on hard prompts, outperforming AR models LlamaGen by 0.18 and diffusion models LDM by 0.15. Exhaustive large-scale human evaluations also demonstrate our prominent image generation ability in terms of text-alignment, visual flaw, and visual appearance. We hope that Token-Shuffle can serve as a foundational design for efficient high-resolution image generation within MLLMs.
title Token-Shuffle: Towards High-Resolution Image Generation with Autoregressive Models
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2504.17789