Unleashing the Potential of Large Language Models for Text-to-Image Generation Through Autoregressive Representation Alignment

Xing Xie; Jiawei Liu; Ziyue Lin; Huijie Fan; Zhi Han; Yandong Tang; Liangqiong Qu

AAAI 2026technical0 citations

Unleashing the Potential of Large Language Models for Text-to-Image Generation Through Autoregressive Representation Alignment

Xing Xie, Jiawei Liu, Ziyue Lin, Huijie Fan, Zhi Han, Yandong Tang, Liangqiong Qu

Abstract

We present Autoregressive Representation Alignment (ARRA), a new training framework that unlocks global-coherent text-to-image generation in autoregressive LLMs without architectural modifications. Different from prior works that require complex architectural redesigns, ARRA aligns LLM

BibTeX

@inproceedings{aaai2026_unleashingthepot,
  title = {Unleashing the Potential of Large Language Models for Text-to-Image Generation Through Autoregressive Representation Alignment},
  author = {Xing Xie and Jiawei Liu and Ziyue Lin and Huijie Fan and Zhi Han and Yandong Tang and Liangqiong Qu},
  booktitle = {AAAI 2026},
  year = {2026}
}