Scenario Adds Hosted 3D Gaussian Splat Generation With Tencent HY World

Michael Rubloff
Tags

Scenario has added HY World Multi-view to Splat, a Tencent model that reconstructs a 3D Gaussian splat scene from 2 to 64 overlapping photos.
A single video walkthrough substitutes for the photo set, and the model needs no camera rig and no calibration step. Inference resolution runs up to 1920 and splat density up to 15M points. The model is addressed in the app as model_hunyuan-world-multiview-to-splat, and its tag list unfortunately carries a geoRestriction value of EU, US and KR.
The generator joins a catalog that now carries several routes to a splat. HY World Skybox to Splat converts a panorama, and HY World Image to Skybox produces that panorama. The Web App v2.20.0 name three engines behind Scenario's 3D splatting, World Labs Marble, HunyuanWorld, and VAST AI TripoSplat. Marble launched as a generative world model and later reached 1.1 and 1.1 Plus, while TripoSplat arrived as a native ComfyUI splat node in June.
Compute v1.4.0 credited Hunyuan-World 2.0 and World Labs Marble for 3D world generation and panorama to splat conversion, and added automatic camera perspective configuration so TripoSplat output and large interior splats preview immediately. The underlying HY-World 2.0 model comes from Tencent Hunyuan, whose paper describes a four stage pipeline of HY-Pano 2.0, WorldNav trajectory planning, WorldStereo 2.0 world expansion and WorldMirror 2.0 composition.
Tencent's other entries on the platform run mostly to mesh work, with Hunyuan 3D 3.1 Pro in base, multiview and sketch variants, Hunyuan Polygen 1.5, Hunyuan 3D Part, Tencent UV Unwrapping and Tencent Texture Edit.
There is currently no information on credit cost of a splat generation or the pricing tier that carries these models. A studio already running Scenario for 2D asset generation under SOC 2 Type II and single sign on can now issue a scene reconstruction against the same account and the same audit trail.
HY World Multi-view to Splat is available now.
New to Gaussian Splatting? Start here





