AI Tools.

Search
image to 3d by microsoft

TRELLIS.2-4B

TRELLIS.2-4B converts single images into structured 3D assets using a sparse volumetric representation. It outputs meshes and radiance fields simultaneously, making it useful for games and 3D pipelines that need both geometry and appearance. The 4B parameter variant trades some fidelity for substantially lower VRAM requirements compared to the larger TRELLIS variants.

Summary text generated by an automated pipeline from the model card · Not individually reviewed or run by us · How this page is made

Share: X Reddit HN LinkedIn

From the model card

Fields below are copied from the tags and counters on the HuggingFace repository microsoft/TRELLIS.2-4B at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.

Publisher (HF namespace)
microsoft
Pipeline tag
image-to-3d
License tag
mit — read the license file in the repo before relying on it
Language tags
English (en)
Papers cited
arXiv:2512.14692
Downloads (HF counter at last fetch)
1,718,290
Likes (HF counter at last fetch)
1,165
Model card
https://huggingface.co/microsoft/TRELLIS.2-4B

Use cases

  • Converting product photos to 3D models for e-commerce
  • Generating game assets from concept art
  • Creating 3D scene backgrounds from reference images
  • Prototyping AR objects from single photos
  • Feeding 3D geometry into downstream NeRF pipelines

Pros

  • Produces mesh and radiance field outputs from one pass
  • Sparse volumetric representation reduces memory vs dense voxels
  • MIT license permits commercial use without royalties
  • Handles diverse object categories without category-specific tuning

Cons

  • 4B size still requires a capable GPU; not suitable for CPU inference
  • Single-image input means occluded surfaces are hallucinated
  • Mesh quality degrades on thin structures and fine details
  • No official fine-tuning recipe released at launch

Tags

trellis2image-to-3denarxiv:2512.14692license:mitregion:us