Skip to content

GRM

Reconstruct 3D assets from sparse views in about 0.1 seconds

Visit official site

What GRM is

The model is described as a large feed-forward reconstruction model that recovers 3D assets from sparse-view images and represents scenes with dense 3D Gaussians.

What you can do with it

  • Recover a 3D asset from a few views
  • Fuse multi-view pixels into Gaussians
  • Rebuild Gaussians and meshes from named sources

Model notes

  • A Transformer fuses the multi-view inputs
  • Pixel-aligned Gaussians feed dense 3D Gaussians
  • Text-to-3D and image-to-3D are listed as generation potential

Pros & cons

✓ What we like

  • Multi-view information is fused by a Transformer
  • Both 3D Gaussians and meshes can be reconstructed

! What to watch out for

  • Generation tasks are described as potential when paired with diffusion models
  • The reported results rely on sparse-view inputs

FAQ

How fast is reconstruction?

The material states about 0.1 seconds for reconstruction and under 8 seconds for generation.

What sources are named?

The feature list names Zero123++, Instant3D, V3D and SV3D.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review