Stable Fast 3D
An image-to-3D AI model you can run locally on a GPU or CPU, with GLB export, ComfyUI support and hosted access through Stability AI.
Stable Fast 3D turns a single object image into a textured 3D mesh on your own hardware. It's aimed at game and VR developers, designers and people creating product models for e-commerce. Built on TripoSR, it uses a retrained model designed to produce meshes and textures suitable for use in games and other 3D projects.
The output includes UV unwrapping, material parameters and normal maps. It reduces lighting baked into texture colors, so the original image's illumination has less influence on the asset's appearance. Optional triangle or quad remeshing lets you adjust mesh structure, and GLB export makes the result usable in other 3D software. ComfyUI nodes let you include reconstruction in an image workflow.
Local inference supports Linux with CUDA, and a CPU backend works when no GPU is available. Apple Silicon MPS and Windows support are experimental. The MPS backend recommends at least 32GB of unified memory. Stability AI reports generation in about half a second on a GPU with 7GB of VRAM. Access to the model weights requires a Hugging Face account and approval. The code and weights are available under the Stability AI Community License, which permits non-commercial use and commercial use subject to eligibility restrictions; other commercial use requires an enterprise license.
Stability AI also offers hosted access through its API and Stable Assistant. Those services process images in the cloud, while the downloadable model runs locally. Stable Assistant includes a 3D viewer and AR viewing on WebXR-compatible devices.
