LongTextAR

Summary

We present the first model LongTextAR specifically designed for long-text image generation, addressing a significant gap in existing text-to-image methods that typically handle only short sentences.
We pinpoint weak tokenization as a critical barrier for effective text rendering in existing multimodal autoregressive models, such as Chameleon.
offers customizable text rendering with control over font attributes while generalizing to natural image generation through co-training. Our experiments demonstrate its potential for applications like document generation and PowerPoint editing.

Word Accuracy Comparison

Overview

Our trained text-focused tokenizer converts the long-text image into discrete token IDs. A corresponding long-text prompt is generated, and the model is then tasked with predicting the image token IDs based on this long text prompt.

Control over font attributes

LongTextAR have good control over font attributes. (Middle Row is Ground-truth)

The long-text generation ability is highly improved with LongTextAR.

Applications

BibTeX

      
        @article{wang2025beyond,
          title={Beyond Words: Advancing Long-Text Image Generation via Multimodal Autoregressive Models},
          author={Wang, Alex Jinpeng and Li, Linjie and Yang, Zhengyuan and Wang, Lijuan and Li, Min},
          journal={arXiv preprint arXiv:2503.20198},
          year={2025}
        }