Stable Fast 3D: 3D generator — what it does and what you need to run it

Stable Fast 3D is an open-source 3D generator from Stability AI that solves the image-to-3D task. The tool reconstructs a 3D mesh from one or more images, performing UV unwrapping and separating lighting from geometry.

What it does

• Reconstructs a 3D model from one or more images (image-to-3D).
• Performs UV unwrapping and illumination disentanglement — as described by the authors.
• Saves the result in GLB format.

The authors do not specify other export formats.

What you need to run it

• Platform: CUDA or MPS (as described by the author).
• VRAM: about 6 GB for a single image (as described by the author).
• System RAM: the authors recommend running the CPU version if the system has less than 32 GB of unified memory — this is a developer statement, not a measurement.
• Language: Python.
• Model weights files: 3.7 GB (single file).
• Code license: non-standard, requires manual review.
• Weights license: other.

The code is available in the repository at github.com/Stability-AI/stable-fast-3d, and the model weights are at huggingface.co/stabilityai/stable-fast-3d. Project website: stable-fast-3d.github.io.

Who it is for

The tool is designed for developers and technical specialists who need fast mesh reconstruction with UV unwrapping from photos. It suits those working in a Python environment with a GPU that supports CUDA or MPS and has sufficient VRAM. It is not suitable for users without command-line experience or for those whose hardware does not meet the memory requirements. Also note the non-standard code license — it must be reviewed separately before commercial use. Stable Fast 3D occupies its own niche among 3D generators: it does not require a cloud service, but it places significant demands on local hardware. The latest code change dates to January 2025, so the project is in an active development phase.