Ying Yang^1*, Zhengyao Lv^2*, Tianlin Pan^1,3, Haofan Wang⁴, Binxin Yang⁵, Hubery Yin⁵, Chen Li^5†, Chenyang Si^1†

¹PRLab, Nanjing University   ·   ²The University of Hong Kong   ·   ³UCAS   ·   ⁴Lovart AI   ·   ⁵WeChat, Tencent Inc.

* Equal Contribution     † Corresponding Author

• Paper released
• Demo video released
• Code coming soon

In bitwise autoregressive models, the predicted probabilities often become overly peaked, giving one class near-certain confidence.
This causes top-p sampling to collapse into a deterministic choice, removing randomness.
As shown in the two subplots below, this collapse of bit-level randomness restricts feature variation, and diversity degradation mainly arises from
(1) the binary classification nature of bitwise modeling and
(2) the overconfident output distributions.

※ The diversity issue mainly originates from bitwise tokenization, rather than from VAR (Visual Autoregressive Modeling) itself.

DiverseAR is an effective approach that enhances image and video diversity in bitwise autoregressive modeling without sacrificing visual quality.
It introduces adaptive logits smoothing and an energy-based generation path selection strategy to achieve richer, more diverse sampling.

If you find this work helpful, please consider citing:

@article{yang2025diversear,
  title={DiverseAR: Boosting Diversity in Bitwise Autoregressive Image Generation},
  author={Yang, Ying and Lv, Zhengyao and Pan, Tianlin and Wang, Haofan and Yang, Binxin and Yin, Hubery and Li, Chen and Si, Chenyang},
  journal={arXiv preprint arXiv:2512.02931},
  year={2025}
}