How VAST Became the Centerpiece of SIGGRAPH 2026: A Chinese AI 3D Pioneer

From July 19‑23, 2026, SIGGRAPH in Los Angeles gathered nearly ten thousand participants from 69 countries to showcase the latest in rendering, geometry modeling, animation, and generative AI. Among the four keynote speakers—Disney, NVIDIA, Bolt Graphics, and VAST—China’s VAST stood out as one of the most frequently mentioned Chinese teams.

VAST’s chief scientist Dr. Cao Yanpei delivered a keynote titled The Teapot Test: First It Tested Showing. Now It Tests Making . He recalled the historic 1975 Newell teapot, a benchmark for rendering realism, and argued that in the generative‑AI era the test must evolve: a model should not only display a teapot but also create a physically usable object that can be placed in games, 3D‑printed, or used to train robots.

In the Real‑Time Live! competition, VAST won the Best in Show award. Their demo, Create Interactive 3D Assets in Seconds! , generated fully textured, rigged, and animated 3D assets from user prompts in real time, placed them into a shared virtual world, and even allowed audience‑driven gameplay where generated props affected a racing game.

Beyond the main stage, VAST co‑hosted a hackathon called Worlds in Action with Li Fei‑Fei’s World Labs, attracting over 500 creators who built next‑generation AI‑native 3D interaction prototypes. A subsequent industry mixer gathered about 200 professionals from AI platforms, visual‑effects studios, and virtual production teams to discuss the future of AI‑driven 3D content.

During the same week, Dr. Cao Yanpei was named to MIT Technology Review’s 2025 TR35 China list, highlighting his continuous contributions from 3D reconstruction and neural rendering to AI‑generated, animatable 3D assets.

VAST’s technical depth was demonstrated by five papers accepted to SIGGRAPH’s Technical Papers track, covering mesh generation, 3D Gaussians, texture synthesis, animatable asset generation, and topology‑agnostic motion priors. Highlights include:

Nexus: Native Mesh Generation with Diffusion – introduces a diffusion‑based approach that decouples vertex and topology generation, enabling the Tripo P1.0 model to output clean, game‑engine‑ready meshes in under two seconds.

Generative 3D Gaussians with Learned Density Control (DeG) – proposes a density‑sampled Gaussian method that makes 3D Gaussian generation fully differentiable, powering the open‑source TripoSplat project.

PixTex: Consistent 3D Texturing via Pixel‑Space Multi‑View Diffusion – solves multi‑view texture seams and blurring, delivering game‑ and film‑ready textures.

AniGen: Unified S³ Fields for Animatable 3D Asset Generation – unifies shape, skeleton, and skinning generation in a shared space, outperforming baselines on topology correctness and skinning quality across diverse categories.

TopoCap: Learning Topology‑Agnostic Motion Priors for Monocular Video‑to‑Animation – introduces the first topology‑agnostic motion extraction framework, enabling zero‑shot transfer of video motion to arbitrary skeletal rigs and releasing a 5,000‑plus skeleton dataset (MOBJAVERSE).

These papers form a coherent pipeline: Nexus underpins Tripo P1.0, DeG fuels TripoSplat, while PixTex, AniGen, and TopoCap fill the remaining gaps of texture, rigging, and animation. The research, productization, and open‑source releases create a closed loop that validates VAST’s claims.

Reflecting on the new "Teapot Test," the article argues that the next benchmark for AI‑3D companies is whether generated assets can truly enter the world—be it games, 3D printers, or robotic simulators. VAST’s progression from a 2023 keynote on lowering modeling barriers to a 2026 showcase of end‑to‑end, interactive 3D creation illustrates China’s rising influence in high‑end computer graphics and generative AI.