Shap-E is OpenAI’s generative model for creating 3D objects from text prompts or images. Released as open source under the MIT license, the project has collected more than 12,200 stars on GitHub, with its last updates in 2024. It is the successor to Point-E and remains one of the most accessible ways to experiment with text-to-3D generation locally.
Instead of generating point clouds or meshes directly, Shap-E learns to produce the parameters of implicit functions that can then be rendered as textured meshes or neural radiance fields (NeRFs). This two-stage design — an encoder that maps 3D assets into a latent representation, and a diffusion model conditioned on text or an image — yields cleaner, more coherent objects than earlier approaches, and generation takes seconds rather than the minutes that optimization-based methods need. The repository includes sampling code, example notebooks and everything needed to render the outputs in Blender or view them as NeRFs.
Shap-E is best understood as a research-grade building block: its outputs are stylized rather than production-detailed, but the approach influenced the wave of text-to-3D work that followed, and the tidy codebase makes it a favorite teaching example for how diffusion models extend beyond images into 3D content.
Shap-E is completely free and open source. Download it from GitHub and run it on your own GPU; there is no hosted paid tier.
Shap-E suits 3D artists, game developers, researchers and hobbyists who want to prototype objects from words, study text-to-3D methods, or generate placeholder assets — keeping expectations realistic about fidelity compared with modern commercial 3D generators.