Vi3W
Skip to content

Guides

Stable Fast 3D explained: Stability AI's fast image-to-3D model

Stable Fast 3D explained: UV-unwrapped image-to-3D from Stability AI, 6 GB VRAM, Mac and Windows notes, install steps and the licence terms.

2 min read

An hourglass of glowing blue sand beside a teapot model assembling itself from floating triangles
Illustration. SF3D is built for speed, with UVs included.

Quick answer

What is Stable Fast 3D?

Stable Fast 3D (SF3D) is Stability AI's open model, released in July 2024, that turns one image into a UV-unwrapped, textured mesh with lighting removed from the texture. It needs about 6 GB of VRAM, has experimental Mac and Windows support, and uses the Stability AI Community Licence, free for organisations under US$1M annual revenue.

What makes SF3D different

The full name says it: "Stable Fast 3D Mesh Reconstruction with UV-unwrapping and Illumination Disentanglement" (GitHub).

  • UV-unwrapped output: the mesh comes with a UV layout, so the texture is a normal image map, not baked vertex colours.
  • Illumination disentanglement: it tries to remove the photo's lighting from the texture, so the model relights properly in Blender or a game engine.
  • Fast and light: about 6 GB VRAM for one image with default settings.
  • GLB output saved to output/.

Platforms

From the README:

  • Linux with CUDA: the main path.
  • Mac (Apple silicon): via the MPS backend, experimental. MPS uses more memory than CUDA; the README recommends the CPU version on systems under 32 GB of unified memory. Tested on an M1 Max with 64 GB.
  • Windows: experimental; needs Visual Studio 2022 and matching PyTorch and CUDA.

Install

  1. Install PyTorch for your platform (match the CUDA version).
  2. pip install -r requirements.txt.
  3. For the Gradio demo: pip install -r requirements-demo.txt.
  4. Run on an image; multiple paths can be passed.

It's also in the ComfyUI 3D Pack. See ComfyUI 3D Pack.

An hourglass beside a teapot assembling from triangles
Illustration. Quick turnaround is SF3D's main appeal.

Licence: read this before commercial use

SF3D uses the Stability AI Community Licence. It allows research, non-commercial and limited commercial use free of charge for people and organisations with annual revenue under US$1,000,000. Commercial users must register with Stability AI, and distributing the model or derivatives requires attribution ("Powered by Stability AI"). Above the revenue threshold you need an enterprise licence. Details: open-source 3D AI licences.

When to use SF3D

  • Quick, clean-UV assets from product-style photos.
  • Mac users wanting something that runs locally (with patience).
  • Pipelines where de-lit textures matter for relighting.

For more detail or stronger shapes, compare with TRELLIS or Hunyuan3D: open-source image to 3D.

Stronger models in the browser

Vi3W runs TRELLIS, TRELLIS 2, Hunyuan 3D and commercial models like Meshy and Tripo with no install, then retopologises and exports GLB, FBX or OBJ. Try Vi3W.

Frequently asked questions

stable fast 3d vs triposr

Both are fast and need about 6 GB. SF3D outputs a UV-unwrapped mesh with de-lit textures, which is more usable as an asset; TripoSR has the simpler MIT licence.

stable fast 3d vs hunyuan3d

Hunyuan3D 2.x is a larger diffusion pipeline with stronger shapes and textures but heavier hardware needs. SF3D is the quick, light option.

how to install stable fast 3d

Install PyTorch for your platform, then pip install -r requirements.txt from the repo, plus requirements-demo.txt for the Gradio demo. Windows needs Visual Studio 2022 for the build.

Last reviewed