SplatsThe evolution of media, in brief
RSS

NVIDIA bets the next era of graphics on world models

An AI agent rigs a 3D character inside Houdini in NVIDIA's SIGGRAPH agentic-AI demo

Image: NVIDIA · Industry

Tessa Marchetti

Tessa Marchetti

Jul 20, 2026-Industry

NVIDIA used its SIGGRAPH 2026 keynote — "Next Era of Graphics: Neural Rendering, World Models, and Simulation" — to lay out a roadmap where captured and generated 3D worlds converge.

Zoom in:

  • The Cosmos 3 world model family comes in three sizes: Edge (4B parameters, optimized for edge deployment), Nano (16B), and Super (64B).
  • Cosmos-Dreams is a collection of closed-loop simulators that generate entire worlds from single frames — running on a single RTX PRO 6000 GPU.
  • ArtiFixer, pitched as a complement to gaussian splats, converts messy real-world 3D captures into clean virtual scenes, fixing noise and missing data, and includes a method for predicting photoreal global illumination directly from scene geometry without ray tracing.
  • Also shown: VideoNeuMat for extracting relightable materials from generative video models, MotionBricks (a real-time motion model trained on 350,000+ motion clips), and DLSS 5's modular three-model neural rendering architecture.

Why it matters: Capture cleanup is the least glamorous, most requested feature in splatting pipelines. When NVIDIA aims research at exactly that pain point — while framing graphics' future around neural rendering and world models — it validates the whole 3D capture stack.

What's next: NVIDIA announced Model Context Protocol integrations spanning Adobe, Blender, SideFX Houdini 22, and Unreal Engine — wiring AI agents directly into the industry's 3D content tools.

Go deeper:

  • NVIDIA Blog: At SIGGRAPH, NVIDIA advances graphics and simulation with agentic and physical AI
  • Wccftech: NVIDIA graphics research accelerates simulation & physical AI at SIGGRAPH
⟵ Back to the brief

More stories

The SuperSplat 3.3.0 editor with a Gaussian splat capture of a street cafe loaded — furled yellow umbrellas over metal tables, parked cars and a tree-lined street behind. The scene manager and transform panels sit at the left, the tool strip along the bottom, and the status bar reads two million splats

SuperSplat rewrote itself on WebGPU and deleted the fallback

Today

Two rows of photoacoustic reconstructions of a branching vascular phantom, shown for SlingBAG, for PAGS, and as the ground-truth digital phantom. The SlingBAG panels carry a mottled noise floor around the vessels; the PAGS panels are cleaner, with the vessel network closer to the crisp white tracery of the phantom

Splatting, but the light is sound and the camera is a transducer

Sep 1, 2026

A schematic of a scene divided into a wireframe grid of cells against black. One cell is outlined in yellow and holds a sharp green cylinder; a blurred blue slab sits behind it and a red slab in front, standing in for the frozen regions flattened into single background and foreground images

Their trick makes VRAM independent of scene size. The test scenes were too small to show it.

Aug 31, 2026

A fairground drop-tower ride rendered twice: on the left from a degraded reconstruction, where the tower and foliage dissolve into white streaks and smears, and on the right after refinement, sharp and photographic against a clear sky

Give it scattered keypoints and it matches a full splat reconstruction

Aug 29, 2026

splats

Short daily briefs on the evolution of media — gaussian splats, volumetric video, dome theaters, headsets, and the research underneath.

Newsroom

  • Latest
  • All stories
  • RSS feed
© 2026 Splats · Terms · Privacy