TripoSR
Open-source model for fast feedforward 3D reconstruction of textured meshes from a single image.
At a Glance
TripoSR model code, pretrained models and demo released under the MIT license; self-hosted, runs locally. Tripo Studio hosted plans are a separate product and not included.
Engagement
Available On
Listed Nov 2026
About TripoSR
TripoSR is an open-source model for feedforward 3D reconstruction from a single image, developed collaboratively by Tripo AI and Stability AI. The code, pretrained models and an online demo are released under the MIT license. The repository states it generates 3D models in under 0.5 seconds on an NVIDIA A100 GPU.
What It Is
TripoSR takes a single input image and reconstructs a 3D object from it. It builds on the Large Reconstruction Model (LRM) approach and adds changes described in its technical report, including channel number optimization, mask supervision and a more efficient crop rendering strategy. Training data comes from a curated CC-BY subset of Objaverse. The project is aimed at researchers, developers and creatives working in 3D generative AI and content creation.
How It Is Used
Users install Python (3.8 or later) and PyTorch, with CUDA if available, then install the requirements. Running run.py on one or more images writes reconstructed 3D models to an output directory. By default the output uses vertex colors; a bake-texture option exports a texture instead, with a configurable texture resolution. A single image uses about 6GB VRAM by default. A local Gradio app is also provided, and a Hugging Face demo is linked.
Performance Claims
The authors report that TripoSR outperforms other open-source alternatives, such as OpenLRM, across multiple public datasets in qualitative and quantitative evaluations. Details are in the arXiv technical report (2403.02151). The Tripo blog says it is suitable for entertainment, gaming, industrial design and architecture, and that it is accessible with or without a GPU.
Community Discussions
Be the first to start a conversation about TripoSR
Share your experience with TripoSR, ask questions, or help others learn from your insights.
Pricing
Open Source (MIT)
TripoSR model code, pretrained models and demo released under the MIT license; self-hosted, runs locally. Tripo Studio hosted plans are a separate product and not included.
- MIT license covering source code, pretrained models and interactive online demo
- Local inference via run.py or local Gradio app
- Default options take about 6GB VRAM for a single image
- Generates 3D models in under 0.5 seconds on an NVIDIA A100 GPU
Capabilities
Key Features
- Single-image feedforward 3D reconstruction
- Textured mesh output in under a second on an NVIDIA A100
- Optional texture baking with configurable texture resolution
- Batch inference over multiple images
- Local Gradio app
- MIT-licensed code and pretrained models
