Abstract
One-dimensional (1D) variable-length visual tokenizers enable adaptive compression by varying the number of tokens, allowing downstream autoregressive (AR) models to flexibly trade off generation quality against computational cost using a single tokenizer. However, existing approaches based on nested dropout often fail to fully exploit the representational capacity of the tokenizer, resulting in suboptimal performance in both image reconstruction and generation. In this work, we introduce NesTok, a nested self-alignment framework tailored to dynamic visual tokenizers. NesTok introduces cross-length training, which jointly optimizes reconstruction across token lengths while using the full-length sequence to guide shorter counterparts, enabling shorter token sequences to approach the reconstruction quality of full-length sequences. On ImageNet, NesTok improves substantially over standard training and achieves an rFID score of 0.98. On downstream image generation, it achieves the state-of-the-art gFID score of 1.46 on ImageNet 256$\times$256 among existing variable-length autoregressive image generation methods. Code will be available at https://github.com/Jiawei804/NesTok.
Keywords
Subject
Publication details
- Journal
- Not available
- Open access
- Green open access
Cite this article
APA 7
Zhang, J., Liu, S., Huang, R., Li, Y., Li, Z., Chang, X., & Li, C. (2026). NesTok: Nested Self-Aligned 1D Tokenizer for Autoregressive Image Generation. https://omanscience.com/en/articles/nestok-nested-self-aligned-1d-tokenizer-for-autoregressive-image-generation
MLA 9
Zhang, Jiawei, et al. "NesTok: Nested Self-Aligned 1D Tokenizer for Autoregressive Image Generation." https://omanscience.com/en/articles/nestok-nested-self-aligned-1d-tokenizer-for-autoregressive-image-generation.
Chicago (author–date)
Zhang, Jiawei, Shuhao Liu, Rong Huang, Yuancheng Li, Zhihui Li, Xiaojun Chang, and Changlin Li. 2026. "NesTok: Nested Self-Aligned 1D Tokenizer for Autoregressive Image Generation." https://omanscience.com/en/articles/nestok-nested-self-aligned-1d-tokenizer-for-autoregressive-image-generation.
Harvard
Zhang, J., Liu, S., Huang, R., Li, Y., Li, Z., Chang, X. and Li, C. (2026) 'NesTok: Nested Self-Aligned 1D Tokenizer for Autoregressive Image Generation', Available at: https://omanscience.com/en/articles/nestok-nested-self-aligned-1d-tokenizer-for-autoregressive-image-generation.
Vancouver
Zhang J, Liu S, Huang R, Li Y, Li Z, Chang X, et al. NesTok: Nested Self-Aligned 1D Tokenizer for Autoregressive Image Generation. https://omanscience.com/en/articles/nestok-nested-self-aligned-1d-tokenizer-for-autoregressive-image-generation
IEEE
J. Zhang, S. Liu, R. Huang, Y. Li, Z. Li, X. Chang, and C. Li, "NesTok: Nested Self-Aligned 1D Tokenizer for Autoregressive Image Generation," https://omanscience.com/en/articles/nestok-nested-self-aligned-1d-tokenizer-for-autoregressive-image-generation.