1 paper
Chunliang Li, Tianze Cao, Sanyuan Zhao
Visual Autoregressive (VAR) modeling inefficiently applies a fixed computational depth to each position when generating high-resolution images. While existing methods accelerate in…