# Shap-E — card for AI agents

OpenAI's text-to-3D and image-to-3D model that outputs textured meshes or neural radiance fields.

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

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

## License
MIT — https://github.com/openai/shap-e/blob/main/LICENSE

- commercial: **conditions** (Commercial use with conditions)
  > we advise against employing them for commercial purposes at this time
  (https://github.com/openai/shap-e/blob/main/model-card.md)
- attribution: **no** (No credit needed)
  > The above copyright notice and this permission notice shall be included in all
  (https://github.com/openai/shap-e/blob/main/LICENSE)
- in app: **yes** (Can ship inside your app)
  > in the Software without restriction, including without limitation the rights
  (https://github.com/openai/shap-e/blob/main/LICENSE)

## Rules to follow
- Keep the MIT license file with the code or weights you ship.
- OpenAI's model card advises against commercial use for now because of quality limits and biases.

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

## Notes
Hugging Face API, Oct 25, 2026: 8,606 downloads, 296 likes; GitHub openai/shap-e: 12,265 stars.

## Numbers (AgentGid data)
- Gid Score: 35
- GitHub stars: 12,267

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

```
- Shap-E (AI 3D model generators). OpenAI's text-to-3D and image-to-3D model that outputs textured meshes or neural radiance fields. Price: free; commercial use with conditions; no credit needed.
  Connect: download from https://huggingface.co/openai/shap-e (no API).
  Rule: Keep the MIT license file with the code or weights you ship.
  Rule: OpenAI's model card advises against commercial use for now because of quality limits and biases.
  License: https://github.com/openai/shap-e/blob/main/LICENSE (checked 2026-11-10; details: https://agentgid.com/assets/shap-e/)
```

Not verified: The Hugging Face card under the openai account was written by the diffusers team, not the Shap-E authors; license facts were taken from the official GitHub repo.

## Links
- AgentGid page: https://agentgid.com/assets/shap-e/
- Is it free for commercial use? https://agentgid.com/assets/shap-e/commercial-use/
- Source: https://huggingface.co/openai/shap-e
- Source: https://github.com/openai/shap-e
- Source: https://github.com/openai/shap-e/blob/main/LICENSE
- Source: https://github.com/openai/shap-e/blob/main/model-card.md
- Source: https://huggingface.co/api/models/openai/shap-e


## Shap-E: rough 3D objects from a text or picture

Shap-E is a free OpenAI research model that turns a short text, like "a birthday cupcake", or a simple picture into a 3D object. You get a spinning preview or a 3D file you can open in other tools. Results are rough and cartoon-like, so it suits quick drafts and experiments rather than finished models.

### Where to use it
- Quick 3D drafts from a sentence: Type what you want, such as "a green boot", and get several rough 3D versions to compare before anyone models the real thing.
- A 3D file for web viewers and games: Save the result as a 3D file, then convert it to the common GLB format that 3D web viewers and game engines open.
- A 3D object from one picture: Start from a picture of a single object on a plain background and get a 3D version of it. Remove the background first for a better result.
- Shapes for 3D printing tests: Make a first rough shape for a small 3D print, then check and fix it in a modeling tool before printing anything that must hold weight.
- Placeholder props in a game prototype: Fill an early game level or a 3D scene with rough cartoon props, like a cupcake or a spaceship, until real models are ready. (idea, not from the official pages)

### How to start
- Install the Hugging Face libraries in a Python environment (a GPU is much faster, but the docs also show a CPU fallback).

```
#!pip install -q diffusers transformers accelerate trimesh
```

- Load the text model and generate frames for one or more prompts.

```
pipe = ShapEPipeline.from_pretrained("openai/shap-e", dtype=torch.float16, variant="fp16")
```

- Save a spinning preview with export_to_gif, or ask for output_type="mesh" and save a 3D file.

```
export_to_gif(images[0], "firecracker_3d.gif")
```


### License
The code and weights are under the MIT license, which allows paid use if you keep the license notice. OpenAI's own model card still advises against commercial use for now because of quality limits and biases.

### Good to know
- Results are rough. Expect holes, rough edges and blurry surfaces, and a simple cartoon style from the training data.
- It struggles with scenes of several objects, with tying colors to the right object, and with exact counts.
- The picture-based model was only checked on clean computer-made renders, so real photos may give worse shapes.
- The project is a 2023 research release and its code has not changed since 2023.
- The Hugging Face page under the OpenAI name was written by the Diffusers team, and its code sample has bugs; use the Diffusers documentation instead.

Checked 2026-11-10 against: https://huggingface.co/openai/shap-e, https://huggingface.co/openai/shap-e/raw/main/README.md, https://huggingface.co/openai/shap-e-img2img/raw/main/README.md, https://huggingface.co/api/models/openai/shap-e, https://huggingface.co/api/models/openai/shap-e/tree/main?recursive=1, https://github.com/openai/shap-e/blob/main/README.md, https://github.com/openai/shap-e/blob/main/model-card.md, https://github.com/openai/shap-e/blob/main/LICENSE, https://github.com/openai/shap-e/blob/main/setup.py, https://github.com/openai/shap-e/blob/main/shap_e/examples/sample_text_to_3d.ipynb, https://github.com/openai/shap-e/commits/main.atom, https://github.com/huggingface/diffusers/blob/main/docs/source/en/api/pipelines/shap_e.md

### Mistakes to avoid
- Don't: Copying the sample from the openai/shap-e Hugging Face page as is (ckpt_id = "openai/shap-e" then from_pretrained(repo)) → Do: Use the Diffusers docs sample, ShapEPipeline.from_pretrained("openai/shap-e", dtype=torch.float16, variant="fp16") (The page defines ckpt_id but loads repo, which is not defined, and installs diffusers from an old branch.)
- Don't: Exporting the mesh straight to GLB and wondering why it is shown from below → Do: Rotate it -90 degrees around the X axis with trimesh before exporting (The mesh comes out viewed from the bottom by default.)
- Don't: Feeding a busy photo to the image model → Do: Cut the object out onto a plain background first (The image model works best on a clean single object.)
- Don't: Using the original repo's encode_model notebook without Blender → Do: Install Blender 3.3.1 or higher and set BLENDER_PATH, or stay with text and image generation (Encoding existing 3D models renders them with Blender.)

### Questions
- Can I sell 3D models made with Shap-E? The MIT license allows it with the notice kept, but OpenAI's model card advises against commercial use for now because of quality limits and biases.
- What file types can I get? A spinning GIF preview, or a mesh saved as PLY or OBJ, which trimesh can turn into GLB.
- Which models are there? openai/shap-e works from text and openai/shap-e-img2img works from a single computer-made image.
- Can it make a whole scene, such as a table with three cups? Usually not well. It often fails when several objects must be placed together or a number is requested.

---
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.
