Right now OpenAI it's not working on ChatGPT5, but is focused on developing other generative models. These days the company presented Shap-E, a new tool that allows you to create 3D models starting from simple text strings. Exceptional results cannot be expected right now, but it is undoubtedly a first step towards a more complex use. Let's see everything in detail.
What is Shap-E how it works
SHAP-E is freely available on GitHub and allows the creation of three-dimensional objects openable on Paint 3D and convertible into STL files, usable also with 3D printers. The Artificial Intelligence in question is based on a new technique of visual synthesis called NeRF (Neural Radiance Fields) which has already attracted great interest in computer vision.
This new method of visual synthesis and 3D reconstruction can be applied to various fields, including robotics (as we recently saw with ChatGPT and the Boston Dynamics dog), cartography, virtual and augmented reality and many others.
In the cases presented by OpenAI, Shap-E is able to generate objects such as a food bowl, a penguin, a dog, a spaceship, but also a chair and a green boot. Models can be rendered in seconds, and as evidenced by many, SHAP-E could be used for creating assets for video games, movies and VR experiences. This is a major turning point in 3D modeling and rendering.
However, Sam Altman's organization did not provide precise instructions on using the tool, but once installed with the Python command "pip install -e", you can use one of the notebooks present on GitHub as "text-to- 3d”, which generates a 3D model from a text prompt, or “image-to-3d”, which transforms a 2D image into a 3D object.
The results aren't great, especially when using text prompts: so you mostly generate monochrome PLY files and not very convincing animated color GIFs. Also, SHAP-E seems to require a lot of system resources.
