# Point-E — card for AI agents

Official code and models for generating 3D point clouds from text or images, plus a model that turns point clouds into meshes.

Category: AI 3D model generators · maker OpenAI · checked 2026-11-10

## Price
- Price: Free
- Free tier: Free open-source code and models under the MIT License

## License
MIT License — https://github.com/openai/point-e/blob/HEAD/LICENSE

- commercial: **conditions** (Commercial use with conditions)
  > Due to the limitations and biases of our models, we do not currently recommend it for commercial use.
  (https://github.com/openai/point-e/blob/main/model-card.md)
- attribution: **no** (No credit needed)
  > The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
  (https://github.com/openai/point-e/blob/HEAD/LICENSE)
- in app: **yes** (Can ship inside your app)
  > to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
  (https://github.com/openai/point-e/blob/HEAD/LICENSE)

## Rules to follow
- Keep the MIT license and copyright notice in all copies or distributions of the code.
- Install locally with pip install -e . and run the models yourself; there is no hosted API.
- The MIT license covers the software; the pages do not state terms for the generated outputs, so check before relying on them.
- OpenAI's model card does not recommend commercial use for now because of quality limits and biases.

## How to connect
- Download only
- Repository: https://github.com/openai/point-e
- Website: https://github.com/openai/point-e

## Notes
Added automatically. License read from the official pages.

## Numbers (AgentGid data)
- Gid Score: 29
- GitHub stars: 6,903

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

```
- Point-E (AI 3D model generators). Official code and models for generating 3D point clouds from text or images, plus a model that turns point clouds into meshes. Price: free; commercial use with conditions; no credit needed.
  Connect: download from https://github.com/openai/point-e (no API).
  Rule: Keep the MIT license and copyright notice in all copies or distributions of the code.
  Rule: Install locally with pip install -e . and run the models yourself; there is no hosted API.
  Rule: The MIT license covers the software; the pages do not state terms for the generated outputs, so check before relying on them.
  Rule: OpenAI's model card does not recommend commercial use for now because of quality limits and biases.
  License: https://github.com/openai/point-e/blob/HEAD/LICENSE (checked 2026-11-10; details: https://agentgid.com/assets/point-e/)
```

## Links
- AgentGid page: https://agentgid.com/assets/point-e/
- Is it free for commercial use? https://agentgid.com/assets/point-e/commercial-use/
- Source: https://github.com/openai/point-e/blob/HEAD/LICENSE
- Source: https://github.com/openai/point-e


## Point-E: rough 3D shapes from text or a picture

Point-E is OpenAI's free research code and models from 2022 that make a rough 3D shape from a short text or a picture. The shape comes out as a cloud of colored dots, and a second small model can turn it into a 3D mesh. The code is MIT licensed, but OpenAI does not recommend it for commercial use.

### Where to use it
- A 3D shape from a few words: Type a simple description such as "a red motorcycle" and get a small colored 3D dot shape of it. It works best with simple objects and colors.
- A 3D shape from a picture: Give it one rendered picture of an object and get a 3D dot shape that matches it, as a starting point for a model.
- Turn the dots into a mesh file: Convert the dot shape into a mesh file you can open in other 3D programs, at low or higher detail.
- Quick prototypes and print sketches: Get a rough first shape of an idea for a prototype or a 3D print sketch, then check and fix it before any real use.
- A baseline for 3D research: Compare a newer 3D model or your own experiment against a known, open starting point.

### How to start
- Get the code from GitHub and install it from the project folder; there is no ready package to install by name.

```
pip install -e .
```

- Open the example notebooks for picture to dots, text to dots, and dots to mesh.
- In the text notebook, pick the small text model; its weights download on first use.

```
base_name = 'base40M-textvec'
```


### License
The code is under the MIT license, so it may legally be used and changed in any project if the notice stays with it. OpenAI's own model card says it does not recommend the models for commercial use.

### Good to know
- Results are rough. The shapes have few points and often show noise, stray dots or cracks.
- The text model only handles simple descriptions; long or unusual requests fail.
- Because it learned from many simple 3D models, results tend to look cartoon-like.
- The model files are downloaded from an OpenAI file server into a point_e_model_cache folder inside the folder you run it from.
- OpenAI warns against trusting printed parts from it without checking them, and against uses where precision matters.

Checked 2026-11-10 against: https://github.com/openai/point-e, https://github.com/openai/point-e/blob/main/README.md, https://github.com/openai/point-e/blob/main/LICENSE, https://github.com/openai/point-e/blob/main/model-card.md, https://github.com/openai/point-e/blob/main/setup.py, https://github.com/openai/point-e/blob/main/point_e/examples/text2pointcloud.ipynb, https://github.com/openai/point-e/blob/main/point_e/examples/image2pointcloud.ipynb, https://github.com/openai/point-e/blob/main/point_e/examples/pointcloud2mesh.ipynb, https://github.com/openai/point-e/blob/main/point_e/models/download.py

### Mistakes to avoid
- Don't: pip install point-e → Do: Clone github.com/openai/point-e and run pip install -e . inside it (The project is installed from its source folder; pypi.org had no point-e project in our check on 2026-11-10.)
- Don't: Prompting the text model with long, detailed scenes → Do: Use short prompts naming one simple object and a color (The text model fails on complex prompts and unusual objects.)
- Don't: Keeping the example grid_size=32 for a final mesh → Do: Use grid_size=128 for the detail the paper used (32 is a quick, coarse setting in the example notebook.)
- Don't: Running the notebooks from different folders and downloading the models again each time → Do: Pass one cache_dir to load_checkpoint, or always run from the same folder (By default the model cache is created inside the current working folder.)

### Questions
- Can I use it in a paid product? The MIT license allows it if you keep the notice, but OpenAI's model card does not recommend commercial use because of the models' limits and biases.
- Is there an online service or key? No. You install the code yourself and run the example notebooks; the model files download on first use.
- What file do I get at the end? The mesh example writes a PLY file with colors that other 3D programs can import.
- Which picture model should I use? The notebook starts with the small base40M model and says base300M or base1B give better results.

---
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.
