# InstantMesh — card for AI agents

Image-to-3D pipeline: a multiview diffusion model plus a sparse-view reconstructor that outputs a mesh in about 10 seconds.

Category: AI 3D model generators · maker Tencent ARC · checked 2026-10-25

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

## License
Apache-2.0 — https://github.com/TencentARC/InstantMesh/blob/main/LICENSE

- commercial: **yes** (Commercial use OK)
  > grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce
  (https://github.com/TencentARC/InstantMesh/blob/main/LICENSE)
- attribution: **no** (No credit needed)
  > You must give any other recipients of the Work or Derivative Works a copy of this License
  (https://github.com/TencentARC/InstantMesh/blob/main/LICENSE)
- in app: **yes** (Can ship inside your app)
  > publicly display, publicly perform, sublicense, and distribute the Work and such Derivative Works in Source or Object form.
  (https://github.com/TencentARC/InstantMesh/blob/main/LICENSE)

## Rules to follow
- Keep the Apache-2.0 license and notices with the code or weights you ship.
- Mark files you changed when you redistribute modified code.

## How to connect
- Download only
- Repository: https://github.com/TencentARC/InstantMesh
- Website: https://huggingface.co/TencentARC/InstantMesh

## Notes
Hugging Face API, Oct 25, 2026: 21,741 downloads, 344 likes; GitHub TencentARC/InstantMesh: 4,577 stars.

## Numbers (AgentGid data)
- Gid Score: 25
- GitHub stars: 4,622

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

```
- InstantMesh (AI 3D model generators). Image-to-3D pipeline: a multiview diffusion model plus a sparse-view reconstructor that outputs a mesh in about 10 seconds. Price: free; commercial use OK; no credit needed.
  Connect: download from https://huggingface.co/TencentARC/InstantMesh (no API).
  Rule: Keep the Apache-2.0 license and notices with the code or weights you ship.
  Rule: Mark files you changed when you redistribute modified code.
  License: https://github.com/TencentARC/InstantMesh/blob/main/LICENSE (checked 2026-10-25; details: https://agentgid.com/assets/instantmesh/)
```

Not verified: Licenses of the third-party models the pipeline downloads (the multiview diffusion stage) were not checked.

## Links
- AgentGid page: https://agentgid.com/assets/instantmesh/
- Is it free for commercial use? https://agentgid.com/assets/instantmesh/commercial-use/
- Source: https://huggingface.co/TencentARC/InstantMesh
- Source: https://github.com/TencentARC/InstantMesh
- Source: https://github.com/TencentARC/InstantMesh/blob/main/LICENSE
- Source: https://huggingface.co/api/models/TencentARC/InstantMesh


## InstantMesh: a 3D model from one picture

InstantMesh is free research software from Tencent ARC that turns one picture of an object into a 3D model you can open in 3D tools. It first draws the object from several sides, then builds the shape from those views. It runs on your own computer with an NVIDIA graphics card, not as a web service.

### Where to use it
- A 3D model from a photo: Take one picture of an object, such as a toy or a product shot. You get a 3D file of it that you can open and edit in Blender or a game engine.
- Models with a texture image: Get the colors as a separate texture picture instead of colored corners, which most game engines and 3D viewers expect. This takes longer.
- A local try-out page: Start a small web page on your own machine where you drop a picture and look at the 3D result, before you write any scripts.
- Pictures that already have no background: If your picture already has a clear, cut-out background, skip the built-in background removal so it does not cut into the object.
- Rough props for a hobby game: Make quick first versions of props from your own sketches or photos for a student or hobby game, then clean them up by hand. (idea, not from the official pages)
- Research on picture-to-3D: Train or fine-tune the models on your own data to compare methods in a study, using the released training code.

### How to start
- On a computer with an NVIDIA graphics card, create a Python 3.10 environment and install the listed versions of the libraries.

```
conda create --name instantmesh python=3.10
conda activate instantmesh
pip install -U pip
conda install Ninja
conda install cuda -c nvidia/label/cuda-12.1.0
pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cu121
pip install xformers==0.0.22.post7
pip install -r requirements.txt
```

- Run the script on a picture. The model files download by themselves on the first run.

```
python run.py configs/instant-mesh-large.yaml examples/hatsune_miku.png --save_video
```


### License
The InstantMesh code and its own model files are under Apache-2.0, which allows paid use. But the default pipeline also loads a multi-view model whose weights are non-commercial, and a required rendering library is non-commercial too. Check both before any paid use.

### Good to know
- You need a computer with an NVIDIA graphics card. The script is written for such cards and will not run on a regular laptop or a Mac.
- The first stage downloads the Zero123++ model from Hugging Face. Its makers say its weights are for non-commercial use only, while its outputs may be used freely.
- The InstantMesh files include a changed copy of the Zero123++ image model, so the same non-commercial question applies to it.
- The required nvdiffrast library from NVIDIA may only be used for research or evaluation, not for making money.
- The code was last changed in January 2025 and pins older library versions, so install exactly the listed versions.

Checked 2026-11-12 against: https://github.com/TencentARC/InstantMesh, https://github.com/TencentARC/InstantMesh/blob/main/README.md, https://github.com/TencentARC/InstantMesh/blob/main/LICENSE, https://github.com/TencentARC/InstantMesh/blob/main/requirements.txt, https://github.com/TencentARC/InstantMesh/blob/main/run.py, https://github.com/TencentARC/InstantMesh/blob/main/configs/instant-mesh-large.yaml, https://huggingface.co/TencentARC/InstantMesh, https://huggingface.co/TencentARC/InstantMesh/raw/main/README.md, https://huggingface.co/api/models/TencentARC/InstantMesh, https://github.com/SUDO-AI-3D/zero123plus/blob/main/README.md, https://github.com/NVlabs/nvdiffrast/blob/main/LICENSE.txt

### Mistakes to avoid
- Don't: pip install torch (latest) and xformers (latest), then pip install -r requirements.txt → Do: Install torch==2.1.0 with the cu121 index and xformers==0.0.22.post7 exactly as the README lists (The pinned diffusers 0.20.2 and the xformers build are matched to PyTorch 2.1.0.)
- Don't: Expecting a texture image from the default command → Do: Add --export_texmap when the target tool needs a texture map (The default export stores colors on the mesh corners.)
- Don't: Using --export_texmap with an instant-nerf config at default settings and waiting a long time → Do: Lower the iso-surface extraction resolution in the config, or use an instant-mesh config (The UV unwrapping step is slow with the NeRF variants at resolution 256.)
- Don't: Treating the Apache-2.0 label on the model page as clearance for a paid product → Do: Check the Zero123++ weight license (CC-BY-NC 4.0) and the nvdiffrast license before commercial use (The default pipeline loads sudo-ai/zero123plus-v1.2 and needs nvdiffrast, both non-commercial.)

### Questions
- Can I run it on a Mac? Not with the official script. It is set up for NVIDIA graphics cards with CUDA 12.1 or newer.
- What file do I get? An .obj model with colors stored on its corners by default, or with a separate texture image when you add --export_texmap.
- Do I have to download the model files by hand? No. The script downloads them on the first run; you can also place them in the ckpts folder yourself.
- Can I use it in a paid product? The InstantMesh code and files are Apache-2.0, but the default pipeline also relies on non-commercial Zero123++ weights and the non-commercial nvdiffrast library, so check those first.
- Which model variant is used by default? The large mesh variant, instant-mesh-large; other variants are picked with a different config file.

---
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.
