G-STAR: Group-Scaled Token Capacity for Deeply Compressed Autoregressive Generation

A new arXiv paper (2611.13004v1) introduces GSTAR, a spatial-group co-scaling framework for discrete autoregressive image generation that achieves 32x and effective 64x compression while preserving reconstruction, understanding and generation quality. The authors show that under a fixed nominal information budget, group size should scale quadratically with the per-axis spatial compression factor, and pair this with a semi-autoregressive decoder that combines parallel feature computation with lightweight causal modeling. At 1024x1024 resolution, GSTAR generates each image in 2.5 seconds on a single NVIDIA H200, a 20x speedup over prior discrete autoregressive models, and performs competitively with leading models across GenEval, DPG-Bench, PRISM, LongText-Bench and Qwen-Image-Bench.

arXiv:2611.13004v1 Announce Type: new Abstract: Discrete autoregressive models offer a scalable framework for image generation. However, they face a fundamental trade-off: long token sequences preserve representation quality but slow generation, while aggressive spatial compression reduces representation capacity. Based on product quantization, we show that, under a fixed nominal information budget, group size should scale quadratically with the per-axis spatial compression factor. Building on this principle, we propose GSTAR, a spatial--group co-scaling framework that enables $32\times$ and even effective $64\times$ compression while preserving strong reconstruction, understanding, and generation performance. Since larger groups introduce intra-group dependencies that must be modeled efficiently, we analyze their impact on inference latency and generation quality and introduce a semi-autoregressive decoder that combines parallel feature computation with lightweight causal modeling. With these designs, our highly compressed models perform competitively with leading models across five representative text-to-image benchmarks GenEval, DPG-Bench, PRISM, LongText-Bench, and Qwen-Image-Bench . At $1024\times1024$ resolution, our model delivers a $20\times$ speedup over prior discrete autoregressive models, generating each image in 2.5 seconds on a single NVIDIA H200. Together, these results show that highly compressed discrete representations can enable high-quality autoregressive generation at low latency.