Shap-e

Shap-E is OpenAI's research model for generating 3D objects conditioned on text or images. The official code and model release is the openai/shap-e repository, licensed under MIT, and the method is described in the paper Shap-E: Generating Conditional 3D Implicit Functions. The repository is a code release, not a hosted product.

Key features

• Text-to-3D: sample a 3D object from a text prompt. The repository's examples include "A chair that looks like an avocado".
• Image-to-3D: sample a 3D object from a synthetic-view image. The repository advises removing the background from the input image for better results.
• Model encoding: render a 3D model or trimesh into multiview images and a point cloud, encode it into a latent, and render it back. This needs Blender 3.3.1 or later and a BLENDER_PATH environment variable.
• MIT license: the code and release are open for reuse under the MIT license.

Model specs

The repository links to a model card and sample gallery. It does not state a deprecation, and it shows no archive notice.

Use cases

• Quick 3D concept sketches from a text description.
• Research on 3D generation and implicit representations.
• Prototyping a pipeline before moving to a production 3D service.

Getting started

1. Clone openai/shap-e and run pip install -e ..
2. Open the text-to-3D or image-to-3D example notebook and sample with your own prompt or image.
3. Export the result and inspect it in a 3D viewer. For the encoding example, install Blender and set BLENDER_PATH.

FAQ

What does Shap-E do?

It generates 3D objects from a text prompt or an image.

Is Shap-E open source?

Yes. The official repository uses the MIT license.

Does Shap-E generate images?

No. Its output is 3D objects. For still images, see GPT Image 2.

Is there a hosted 3D service?

Shap-E is a code release. For a product-style 3D workflow, see Hyper3D Rodin.

Alternatives

• Hyper3D Rodin: a hosted 3D generation product.
• GPT Image 2: OpenAI's image generation model for stills.
• DALL-E 3: OpenAI's earlier image model.

Conclusion

Shap-E is a compact, MIT-licensed way to experiment with text-to-3D and image-to-3D generation. Start with the official repository and the paper, and browse more tools in the models category.