# TripoSR — card for AI agents

Fast feed-forward model that reconstructs a 3D mesh from a single image; the makers cite under 0.5 s on an A100 GPU.

Category: AI 3D model generators · maker Stability AI and Tripo AI · checked 2026-10-25

## Price
- Price: Free
- Free tier: Open weights, free to download.

## License
MIT — https://github.com/VAST-AI-Research/TripoSR/blob/main/LICENSE

- commercial: **yes** (Commercial use OK)
  > to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  (https://github.com/VAST-AI-Research/TripoSR/blob/main/LICENSE)
- attribution: **no** (No credit needed)
  > The above copyright notice and this permission notice shall be included in all
  (https://github.com/VAST-AI-Research/TripoSR/blob/main/LICENSE)
- in app: **yes** (Can ship inside your app)
  > in the Software without restriction, including without limitation the rights
  (https://github.com/VAST-AI-Research/TripoSR/blob/main/LICENSE)

## Rules to follow
- Keep the MIT license file with the code or weights you ship.
- Do not create 3D content meant to disturb or offend, or that spreads stereotypes (model card misuse policy).

## How to connect
- Download only
- Repository: https://github.com/VAST-AI-Research/TripoSR
- Website: https://huggingface.co/stabilityai/TripoSR

## Notes
Hugging Face API, Oct 25, 2026: 248,995 downloads, 680 likes; GitHub VAST-AI-Research/TripoSR: 7,219 stars.

## Numbers (AgentGid data)
- Gid Score: 30
- GitHub stars: 7,426

## Lines to paste into your prompt or AGENTS.md

```
- TripoSR (AI 3D model generators). Fast feed-forward model that reconstructs a 3D mesh from a single image; the makers cite under 0.5 s on an A100 GPU. Price: free; commercial use OK; no credit needed.
  Connect: download from https://huggingface.co/stabilityai/TripoSR (no API).
  Rule: Keep the MIT license file with the code or weights you ship.
  Rule: Do not create 3D content meant to disturb or offend, or that spreads stereotypes (model card misuse policy).
  License: https://github.com/VAST-AI-Research/TripoSR/blob/main/LICENSE (checked 2026-10-25; details: https://agentgid.com/assets/triposr/)
```

Not verified: The 0.5-second speed is the GitHub README's claim and was not tested.

## Links
- AgentGid page: https://agentgid.com/assets/triposr/
- Is it free for commercial use? https://agentgid.com/assets/triposr/commercial-use/
- Source: https://huggingface.co/stabilityai/TripoSR
- Source: https://github.com/VAST-AI-Research/TripoSR
- Source: https://github.com/VAST-AI-Research/TripoSR/blob/main/LICENSE
- Source: https://huggingface.co/api/models/stabilityai/TripoSR


## TripoSR: a 3D model from one picture

TripoSR is a free AI model that turns one picture of an object into a 3D model file for 3D tools, games or web pages. It runs on your own computer, best with an NVIDIA graphics card. The code and the trained model are under the MIT license, so paid use is allowed.

### Where to use it
- A 3D model from a product photo: Turn one photo of a shoe, chair or toy into a 3D file you can open in Blender or show on a shop page.
- Many pictures in one run: Give it a list of pictures, such as every item in a small catalog, and it makes one 3D model per picture.
- A local page to try pictures: Start a small web page on your computer where you drop in a picture, adjust the cut-out and download the 3D model.
- Models with a texture image: Ask for a texture picture instead of colors stored on the shape, which many game engines and 3D tools handle better.
- 3D items on a web page: Save the result as a GLB file and show it on a product or portfolio page with a 3D viewer people can turn with the mouse. (idea, not from the official pages)
- Placeholder props for game prototypes: Make quick 3D stand-ins for props from concept sketches or photos while the real models are still being made. (idea, not from the official pages)

### How to start
- Use Python 3.8 or newer, install PyTorch for your computer (with CUDA if you have an NVIDIA card), then the project requirements.

```
pip install -r requirements.txt
```

- Run the script on a picture; the model weights are fetched from Hugging Face the first time.

```
python run.py examples/chair.png --output-dir output/
```

- Or start the local web page and use it in the browser.

```
python gradio_app.py
```


### License
MIT license for the code and the trained model, so you may use it in paid products and change it. Keep the copyright and license notice with copies, and do not use it to make disturbing or offensive 3D content.

### Good to know
- With the default settings one picture needs about 6 GB of graphics card memory.
- Without an NVIDIA card it falls back to the normal processor, which works but is much slower.
- The trained model is downloaded from Hugging Face on first use; you can also point the script to a local copy.
- The makers now point to a newer model, SF3D, for faster results and more game-ready files.
- It was trained on renders of a chosen part of the Objaverse 3D collection, which is under a CC-BY license.

Checked 2026-11-10 against: https://github.com/VAST-AI-Research/TripoSR, https://github.com/VAST-AI-Research/TripoSR/blob/main/README.md, https://github.com/VAST-AI-Research/TripoSR/blob/main/LICENSE, https://github.com/VAST-AI-Research/TripoSR/blob/main/requirements.txt, https://github.com/VAST-AI-Research/TripoSR/blob/main/run.py, https://github.com/VAST-AI-Research/TripoSR/blob/main/gradio_app.py, https://huggingface.co/stabilityai/TripoSR, https://huggingface.co/stabilityai/TripoSR/raw/main/README.md, https://huggingface.co/api/models/stabilityai/TripoSR

### Mistakes to avoid
- Don't: Installing any PyTorch build and then hitting "torchmcubes was not compiled with CUDA support" → Do: Install the PyTorch build whose CUDA major version matches the installed CUDA, upgrade setuptools, then reinstall torchmcubes from GitHub (torchmcubes must be built against the same CUDA as PyTorch.)
- Don't: Running out of graphics card memory and giving up → Do: Pass a smaller --chunk-size (for example 4096), which uses less memory and takes longer (The chunk size controls memory use during shape extraction.)
- Don't: Expecting a texture image in the output by default → Do: Add --bake-texture (and --texture-resolution if needed) (By default colors are stored on the mesh vertices, not in a texture file.)
- Don't: Passing --no-remove-bg with an ordinary photo → Do: Leave background removal on, or prepare a gray background with the object at a sensible size (With that flag the script expects an already prepared picture.)
- Don't: Starting the web page with --share on a work machine → Do: Run python gradio_app.py without --share, or use --listen only inside a trusted network (--share makes the page reachable through Gradio's public site.)

### Questions
- Can I use the 3D models in a paid game or shop? Yes. The MIT license covers the code and the trained model; keep the license notice with copies and follow the model card's misuse rule.
- Which file types does it save? OBJ by default, or GLB when you pass --model-save-format glb.
- Do I have to remove the background first? No. It removes the background by default; with --no-remove-bg you must supply a picture with a gray background and a well-sized object.
- Is there an online version to try first? Yes, the makers link a demo page on Hugging Face; there is no paid service or key in this project.

---
AgentGid's catalog is curated: every entry was checked against the maker's official pages (registry, repository, license and pricing pages) on the date shown; a license value shows the quote it stands on wherever AgentGid has one. Usage numbers come from AgentGid's own scan of public repositories set up for AI coding agents. Claims by makers are reported, not tested by AgentGid.
