VAST at SIGGRAPH: Generating Animatable 3D Assets in Seconds with 5 Papers
机器之心 · wechat · 2026-07-30
At SIGGRAPH 2026, 3D large model company VAST took center stage, winning the top award in the Real-Time Live! segment and securing 5 papers in Technical Papers.
Redefining the "Teapot Test"
In his Keynote, VAST's Chief Scientist Yanpei Cao argued that the generative AI era requires upgrading the "Teapot Test" from "can it realistically render" to "can it truly create." This means generated 3D models must have reliable meshes, editable topologies, and physical properties to be directly used in game engines, 3D printing, or robot training.
Full-Stack Tech Pipeline
Their 5 selected papers cover the entire 3D asset creation pipeline, from mesh generation and texture mapping to skeletal rigging. The core tech behind their flagship Tripo P1.0 abandons sequential generation, outputting clean, low-poly meshes ready for engines in just 2 seconds. Another 3D Gaussian generation tech has been open-sourced and topped HuggingFace's trending list.
Industry Positioning
VAST also co-hosted a hackathon with Fei-Fei Li's World Labs, exploring the integration of 3D large models and world models, signaling that Chinese teams are now core participants in the global AI 3D foundation model space.
More from Multimodal
- MiniMax H3 Tested: Impressive Results with 3x3 Grid Prompts — techhalla · 2026-07-30
- From Single Image to Cinematic Video: Complete AI Workflow Revealed — SimplyAnnisa · 2026-07-30
- Users Anticipate Open-Source FLUX 3 Video Model, Complain About LTX 2.3 Glitches — EverythingMacPro · 2026-07-30
- Sarvam AI Voice Model Demo: Cinematic Expressiveness Shocks Testers — Scobleizer · 2026-07-30
- CivitAI Massively Removes Adult and Face LoRAs, Shocking Returning Users — literally-me-bro · 2026-07-30
- Restoring Black & White Videos with LTX 2.3 AI Colorization Workflow — BrilliantIll5962 · 2026-07-30