NAMVIS: Parallel Multi-View Image Synthesis via Geometry-Conditioned Autoregression
October 2, 2026
NAMVIS replaces iterative diffusion denoising with a diffusion-free, next-scale autoregressive framework for 3D content creation. It uses Multi-scale Projective Pose Encoding to predict discrete visual tokens in parallel across views, significantly reducing inference latency compared to standard diffusion models.
HOW THIS AFFECTS YOU
●
builderYou can achieve faster multi-view generation by moving away from iterative diffusion sampling.
●
researcherThis method offers a new way to anchor autoregressive generation to explicit camera geometry.
●
designerYou can utilize faster 3D content generation workflows for real-time creative applications.