Nvidia Develops ArtiFixer AI Model for 3D Scene Generation From Photos
ArtiFixer addresses a known weakness in 3D Gaussian Splatting by generating and inserting parts not captured in reference photos, producing sharper and more consistent 3D scenes.
Reporting from 1 sources: GIGAZINE.
Nvidia has announced ArtiFixer, an AI model based on the Wan 2.1 video generation AI that can generate 3D scenes from multiple photos. The model fills in missing parts using generative processing, improving on conventional 3D Gaussian Splatting methods that often fail to maintain scene consistency. ArtiFixer comes in three variants: ArtiFixer, ArtiFixer3D, and ArtiFixer3D+.
Nvidia has developed ArtiFixer, an AI model that generates 3D scenes from multiple photos by filling in missing parts using generative processing. The model is based on the Wan 2.1 video generation AI and has approximately 16.9 billion parameters. It was trained in two stages: first to generate and insert parts not in the photos, then distilled into an autoregressive model that generates hundreds of frames from a single frame.
ArtiFixer comes in three variants. ArtiFixer uses an autoregressive model to generate views. ArtiFixer3D distills that output into a 3D representation. ArtiFixer3D+ applies an autoregressive model as post-processing to ArtiFixer3D's results. Nvidia says ArtiFixer3D+ can generate scenes that are both sharp and highly consistent, outperforming conventional methods like 3DGUT, GenFusion, and GSFixer in quality.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.