MaskGIT Sampler Explained: Adaptive Order Selection Speeds Diffusion

New “moment sampler” for MaskGIT matches quality while cutting wall‑clock time; a hybrid adaptive unmasking speeds convergence on image and text benchmarks. Read more: getnews.me/maskgit-sampler-explaine... #maskgit #diffusion

What Drives Compositional Generalization in Visual Generative AI

A new study shows that adding a continuous JEPA loss to the discrete MaskGIT model improves compositional generalization in image and video generation. Read more: getnews.me/what-drives-compositiona... #maskgit #jepa #visualai

Multi-Codebook Speech Generation with Frame‑Stacked Local Transformers

The study, submitted on 23 Sep 2025, shows frame‑stacked parallel decoding with a MaskGIT local transformer speeds up speech generation while keeping quality acceptable. Read more: getnews.me/multi-codebook-speech-ge... #multicodebooks #maskgit