3D catalogue · Verified local path

BuildRome local guide

Non-commercial single-image scene reconstructor that builds metric 3D scene meshes and exports GLB assets.

Choose an app

Start with the recommended app, or choose another supported option.

Compare 3D models
Open model filesRecommended verified starting pointOpen advanced setupPython runtime for experienced usersOpen official sourcePublisher repository and primary evidence

Official example

Play the official source preview directly below. The media stays on the official publisher server and is never hosted by LocalClaw.

What it does

Image To 3dScene ReconstructionMesh ReconstructionNovel View Synthesis

Applied Intuition publishes the BuildRome inference repository plus public Hugging Face delta checkpoints for its TRELLIS.2-based scene generator. The official install targets Linux, Conda, Python 3.10, PyTorch 2.6.0 CUDA 12.4, CUDA compiler extensions, and a separate PyTorch 2.8 MoGe-3 environment; first use downloads TRELLIS.2, TRELLIS image-large, DINOv3 and MoGe-3 dependencies from their upstream sources. The CLI and ScenePipeline API write scene.glb, layout_bands.json, layout_topview.png and spiral.mp4. The README reports a 98 m street-scene run peaking at 15.8 GiB allocated GPU memory on one A100 80GB, so LocalClaw records 64 GB RAM and 24 GB NVIDIA VRAM as a conservative practical floor while noting the workstation CUDA stack.

Hardware figures are practical entry floors, not performance guarantees. Resolution, duration, precision, offloading and runtime versions can materially change memory use.

Strengths

  • Official code, model card and public checkpoint deltas
  • Single-photo metric scene mesh reconstruction
  • Exports glTF/GLB scenes with layout metadata and orbit previews

Limits to know

  • Non-commercial CC BY-NC 4.0 release
  • Depends on several upstream models with their own terms and downloads
  • Linux/CUDA research stack with custom extension builds