Research model that predicts layered, pixel-aligned 3D point clouds, including hidden geometry, from one image or video.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
World Tracing is a research model from a NeurIPS 2026 Spotlight paper (Hao Zhang, Mohamed El Banani, Gengshan Yang and co-authors). For each input pixel it predicts an ordered stack of camera-space 3D points. The first layer is the visible surface, and later layers are front-to-back intersections with occluded surfaces. The project page shows six depth layers per pixel.
The model is a diffusion transformer (WT-DiT) that treats geometry layers as separate denoising tokens, trained with pixel-space flow matching. It handles single objects, indoor and outdoor scenes, and short videos, where it produces per-frame layered point clouds that stay 3D-consistent. According to the authors it outperforms both depth predictors and image-to-3D generators on object, scene and dynamic benchmarks.
Because it keeps the 2D-to-3D correspondence, the output can be lifted into textured meshes by off-the-shelf mesh generators without fine-tuning. It also supports 3D scene editing (add, replace, remove objects) and geometry-guided novel-view video with Wan2.2 VACE. A YouTube creator reports three released models totalling about 6.2 GB. No GitHub or Hugging Face links were available in the provided material, so the release details and license are unconfirmed.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Turns a single image or short video into a layered 3D point cloud including hidden geometry, liftable to textured meshes and usable for 3D edits. Three models released, about 6.2 GB.
Closest neighbours by meaning, not just by category.
Model
Research framework that turns posed images with instance masks and 3D boxes into per-object meshes for cluttered scenes.
Model · World Labs
World Labs' world model that generates and reconstructs explorable 3D worlds from images, text, video or 3D input.
Research
Research method that builds controllable, articulated 3D object models from keypoints, point clouds or depth video.
Model
Research model that unifies 3D scene generation and reconstruction in a single pixel-space diffusion model.
Model
Research code that streams video generation chunk by chunk from a rough 3D proxy, built on Code World Model.
Research
Research framework for globally consistent 3D reconstruction from long RGB videos using chunked priors and test-time adaptation.
