NesTok: Nested Self-Aligned 1D Tokenizer for Autoregressive Image Generation

2026-09-29

AI Summary

This study addresses the underutilized representational capacity of existing variable-length visual tokenizers caused by nested dropout, which limits both reconstruction and generation performance. To overcome this, we propose NesTok, a nested self-alignment framework that introduces a cross-length joint optimization mechanism, leveraging full-length sequences to guide the training of shorter ones. This approach substantially improves reconstruction fidelity with short tokens, enabling adaptive compression and efficient generation. The proposed method achieves a reconstruction FID (rFID) of 0.98 on ImageNet and a generation FID (gFID) of 1.46 at 256×256 resolution, establishing new state-of-the-art performance for variable-length autoregressive image generation.

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/jaiwei804/NesTok.

Problem

• variable-length visual tokenizer
• autoregressive image generation
• nested dropout
• image reconstruction

Innovation

• Nested Self-Alignment
• Variable-Length Tokenizer
• Cross-Length Training
• Autoregressive Image Generation
• Dynamic Visual Tokenizer