G-STAR from PKU and Meituan Speeds Up Autoregressive Image Generation 20x

Researchers from Peking University and Meituan's LongCat team introduced G-STAR, a product-quantization-based method that achieves 32-64x token compression and speeds up discrete autoregressive image generation by 20x, producing images in about 2.5 seconds without sacrificing quality.

2026-11-13 ~ 2026-11-14 · 3 related posts

• G-STAR from PKU & Meituan LongCat: 20x faster autoregressive image generation at 2.5s per image — meituan-longcat · 2026-11-13
• G-STAR paper: 32–64x token compression gives 20x faster autoregressive image generation — _akhaliq · 2026-11-14

1 near-duplicate retellings: pmttyji