Shap-E

Generate 3D Models from Text Prompts

Shap-E can generate 3D models conditioned on text prompts. The official repository includes a notebook (sample_text_to_3d.ipynb) that demonstrates sampling a 3D model from a text description.

Create 3D objects directly from descriptive text.

Generate 3D Models from Images

Shap-E also supports generating 3D models conditioned on a synthetic view image. The sample_image_to_3d.ipynb notebook shows how to sample a 3D model from an input image. For best results, the background should be removed from the input image.

Convert a single image into a 3D model.

Encode Existing 3D Models

The encode_model.ipynb notebook allows loading a 3D model or a trimesh, creating a batch of multiview renders and a point cloud, encoding them into a latent, and rendering it back. This requires Blender version 3.3.1 or higher, with the BLENDER_PATH environment variable set to the Blender executable.

Encode and re-render existing 3D assets.

Official Code and Model Release

Shap-E is the official code and model release for the paper 'Shap-E: Generating Conditional 3D Implicit Functions'. The repository is public and open source under the MIT license.

Access and use the official implementation freely.

Example Outputs

The repository includes a samples.md file with random samples on selected prompts, and the README highlights several examples such as 'A chair that looks like an avocado', 'An airplane that looks like a banana', 'A spaceship', 'A birthday cupcake', and more.

See what the model can generate.

Easy Setup with pip

Shap-E can be installed with the command 'pip install -e .' from the repository. The project is primarily written in Python (98.3%) with some Jupyter Notebook (1.7%).

Get started quickly with a simple pip install.

• Official OpenAI ReleaseShap-E is the official code and model release from OpenAI for generating 3D objects conditioned on text or images, as stated in the repository description.
• Text and Image ConditioningThe repository provides notebooks for text-to-3D (sample_text_to_3d.ipynb) and image-to-3D (sample_image_to_3d.ipynb) generation, with a note that removing background from input images yields best results.
• 3D Model Encoding and RenderingThe encode_model.ipynb notebook loads a 3D model or trimesh, creates multiview renders and a point cloud, encodes them into a latent, and renders it back. This requires Blender 3.3.1 or higher and setting the BLENDER_PATH environment variable.
• Active Community and UsageThe repository has 12.3k stars, 1.1k forks, and 231 watchers, indicating active community interest. It includes sample outputs and a model card, and is licensed under MIT.

• Installation and DependenciesInstallation is via 'pip install -e .' but the repository does not list specific dependencies or Python version requirements. Users must ensure their environment meets unspecified prerequisites.
• Blender Requirement for EncodingThe encode_model.ipynb notebook requires Blender version 3.3.1 or higher and the BLENDER_PATH environment variable. Without this, the encoding functionality will not work.
• Sample Quality and LimitationsThe repository shows highlighted samples but does not guarantee output quality for all prompts. The text-conditional model's performance on arbitrary prompts is not fully documented.
• No Pricing or Commercial TermsThe scraped content does not mention any pricing, licensing costs, or commercial usage terms beyond the MIT license. Users should refer to the full license file for details.

Text-to-3D Generation

Use Shap-E to generate 3D models directly from text prompts, such as 'a chair that looks like an avocado' or 'a spaceship'. The official repository provides a notebook (sample_text_to_3d.ipynb) that demonstrates sampling a 3D model conditioned on a text prompt.

Image-to-3D Generation

Create 3D models from a synthetic view image using the sample_image_to_3d.ipynb notebook. For best results, remove the background from the input image before conditioning.

3D Model Encoding and Rendering

Load an existing 3D model or trimesh, create a batch of multiview renders and a point cloud, encode them into a latent, and render it back. This requires Blender 3.3.1 or higher and setting the BLENDER_PATH environment variable.

Exploring Conditional 3D Implicit Functions

Shap-E is an official code and model release for generating conditional 3D implicit functions. Researchers and developers can use the repository to experiment with the model and see examples of what the text-conditional model can generate.

Sign up for an OpenAI account and obtain an API key from the dashboard.

Install the official SDK (pip install openai‑shap‑e) or use raw HTTP calls.

Test a simple prompt in the sandbox environment to verify response format.

Integrate the API call into your asset pipeline and monitor usage via the OpenAI console.