Nvidia Unveils ArtFixer to Fill 3D Scene Gaps, Targeting 24 GB-Plus Reconstruction Workloads
Updated
Updated · Kotaku · Jun 23
Nvidia Unveils ArtFixer to Fill 3D Scene Gaps, Targeting 24 GB-Plus Reconstruction Workloads
1 articles · Updated · Kotaku · Jun 23
Summary
Nvidia introduced ArtFixer, a research prototype that uses an open autoregressive model to generate missing geometry in 3D scene reconstructions when cameras fail to capture parts of a scene.
The tool is built on 3D Gaussian Splatting, which already speeds reconstruction versus older NeRF methods but struggles in under-observed areas, leaving blank or distorted regions that ArtFixer aims to infer.
Nvidia says the system preserves consistency with observed footage while extrapolating unseen content, with paper, code and a demo released ahead of SIGGRAPH 2026.
The approach could help visual effects, CAD and eventually game asset creation, though current Gaussian-splatting workflows still have limited support for lighting, physics and shadows and typically require at least 24 GB of VRAM.