TripoSR
An open-source image-to-3D model reconstructs objects locally with PyTorch, with MIT-licensed weights and about 6GB VRAM use at default settings.
TripoSR creates a 3D object from a single image on your own hardware. It's for artists, developers and researchers who want image-based reconstruction for their projects or software.
- Single-image reconstruction: Generate an object model from one image, with reported inference under half a second on an NVIDIA A100 GPU.
- Color and texture output: Choose vertex colors or bake a texture for the reconstructed model.
- Batch processing: Process several input images in one run and save the resulting models locally.
TripoSR runs with Python and PyTorch and can use CUDA acceleration where available. Default inference uses about 6GB of VRAM for one image. The reported speed depends on the hardware; the A100 result isn't a timing guarantee for other GPUs. Its source code and pretrained models use the MIT license, and an online demo lets you try reconstruction in a browser.
Tripo's hosted platform is a separate product. Text-to-3D generation, quad meshes, PBR texturing and automatic rigging belong to that online service, rather than the local TripoSR model.
Logotyper
