Guide · Open source
How to use Shap-E: install, run the notebooks and generate 3D models
Updated · written with AI from the project's own documentation and checked against our catalogue data
Shap-E is OpenAI's official research release for generating 3D objects from a text description or from an image. This guide walks through what the repository contains, how to install it from source, and how to run its example notebooks. It is written for readers who are comfortable with Python but are not 3D specialists.
Shap-Eopenai/shap-e · ★ 12.3k · MIT Install options & downloads Quick start
- 1Clone the repository. Download the openai/shap-e source code to your machine and open its folder.
- 2Install the package. Run pip install -e . from inside the cloned repository to install Shap-E in editable mode.
- 3Open the examples. Launch Jupyter and open the notebooks in shap_e/examples.
- 4Run the text-to-3D notebook. Open sample_text_to_3d.ipynb, enter a text prompt and run the cells in order.
- 5Try image input. Use sample_image_to_3d.ipynb with an image whose background you have removed.
- 6Set up Blender if needed. Install Blender 3.3.1 or higher and set BLENDER_PATH before running encode_model.ipynb.
What Shap-E does and what it does not do
Shap-E is the official code and model release for the paper "Shap-E: Generating Conditional 3D Implicit Functions". The repository provides model code and examples in Jupyter notebooks. The README includes sample renders of text-conditioned outputs, such as a chair that looks like an avocado and a penguin.
The README describes research code and notebooks, not an end-user app. There is no graphical interface described in the source, so you work through Python and notebooks.
- Generates 3D objects conditioned on a text prompt or a synthetic view image
- Ships example notebooks rather than a finished product
- Not described as producing production-grade or print-ready meshes
Check the prerequisites before you start
You need Python and pip. Blender is only needed for one notebook. The encode notebook loads a 3D model or trimesh, creates multiview renders and a point cloud, encodes them into a latent, and renders it back.
For that encode notebook, the README asks for Blender version 3.3.1 or higher and an environment variable named BLENDER_PATH that points to the Blender executable. You can skip Blender if you only want the text-to-3D or image-to-3D examples.
- Python
- pip
- Blender 3.3.1 or higher, only for encode_model.ipynb
Install Shap-E from the cloned repository
The README uses a clone-and-build approach. Clone the repository, move into its folder, and install the package in editable mode from that location.
The README gives this install command. It is the only install command the source lists.
pip install -e .Run the text-to-3D notebook
Start with sample_text_to_3d.ipynb. According to the README, it samples a 3D model conditioned on a text prompt. Open the notebook in Jupyter from the repository and run the cells in order.
Text prompts are the main way the model is demonstrated. The README sample prompts include "A spaceship", "A birthday cupcake" and "A green boot", which give a sense of the kind of description that works.
- Notebook: shap_e/examples/sample_text_to_3d.ipynb
- Input: a text prompt
- Output: a sampled 3D model
Run the image-to-3D notebook
The image notebook, sample_image_to_3d.ipynb, samples a 3D model conditioned on a synthetic view image. The README advises removing the background from your input image to get the best result.
Remove the background before you pass the picture in. The source does not name a specific tool for this, so use any image editor you already know.
- Notebook: shap_e/examples/sample_image_to_3d.ipynb
- Remove the image background first for better results
Use the encode notebook only if you need it
encode_model.ipynb loads a 3D model or a trimesh, creates a batch of multiview renders and a point cloud, encodes them into a latent, and renders it back. Set BLENDER_PATH to your Blender executable before running it.
This is the most setup-heavy path. If your goal is simply to see text-to-3D output, you can leave it for later.
- Notebook: shap_e/examples/encode_model.ipynb
- Requires Blender 3.3.1 or higher
- Requires the BLENDER_PATH environment variable
BLENDER_PATH=/path/to/blenderWhat people use it for
- Exploring how text prompts translate into 3D shapes for research
- Trying text-to-3D generation from Python as a developer
- Learning how generative 3D models work by running the notebooks
- Producing quick 3D concept sketches for design exploration
Frequently asked questions
Is Shap-E free to use?
Shap-E is released under the MIT licence, as listed for this repository. The MIT licence generally permits use, modification and redistribution, but read the licence file in the repository for the exact terms.
What do I need to install Shap-E?
You need Python and pip. The README installs the package with pip install -e . from the cloned repository. Blender 3.3.1 or higher is only required for the encode notebook.
Do I need Blender to use Shap-E?
No, not for the basic examples. Blender 3.3.1 or higher is only needed for encode_model.ipynb, which also requires setting the BLENDER_PATH environment variable.
Is there a Shap-E app or web interface?
The README describes research code and notebooks, not an end-user app. You run it through Python and Jupyter notebooks.
How does Shap-E compare to Point-E or DreamFusion?
Point-E, DreamFusion and Shap-E community forks are listed as comparable projects. This source does not include a benchmark comparing them, so test the options yourself on your prompts.
Can I use Shap-E output for 3D printing?
The source does not claim production-quality or print-ready meshes. Treat the output as exploratory and check it carefully before using it for any physical or production work.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Shap-E suits researchers, Python developers, students and designers who want to experiment with text- or image-conditioned 3D generation and are comfortable with notebooks. It is a poor choice if you need a polished end-user tool, a no-code workflow, or print-ready meshes, since the README presents it as research code.
Install options & downloads