Tag
This paper proposes Nemotron-Labs-Diffusion-Image, a masked discrete diffusion model for high-resolution text-to-image synthesis, introducing a token-editing mechanism and grouped cross-entropy objective to improve token refinement and training efficiency.