Autoregressive Visual Generation Needs a Prologue

Benchmark Model Rank Results
image-generation-on-imagenet-256x256Prologue B–B–FID: 4.11
image-generation-on-imagenet-256x256Prologue B–L–FID: 2.67
image-generation-on-imagenet-256x256Prologue B–XL–FID: 2.43
image-generation-on-imagenet-256x256Prologue L–B–FID: 2.15
image-generation-on-imagenet-256x256Prologue L–L–FID: 1.52
image-generation-on-imagenet-256x256Prologue L–XL–FID: 1.46