LongCat team (Xu Huang et al.) · Compressed AR Image

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

G-STAR is a spatial-group co-scaling framework for discrete autoregressive image generation. It enables 32× and effective 64× spatial compression while keeping strong reconstruction, understanding and generation performance, and adds a semi-autoregressive decoder to manage the latency cost of larger groups. The authors report competitive results on GenEval, DPG-Bench, PRISM, LongText-Bench and Qwen-Image-Bench, with a 20× speedup over prior discrete autoregressive models and about 2.5 seconds per 1024×1024 image on a single NVIDIA H200.

Open the source ↗

Similar Vision & Image releases

Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.