From images to chats, from chats to animations, transcriptions, music, videos… yes, artificial intelligence has already reached all these fields, but it still has one box to check: the generation of 3D objects. In September last year, Nvidia told us about its GET3D system. Now it's OpenAI's turn, which after presenting Point-E, decided to publish on GitHub the code for its Shap-E model, with the promise of creating 3D models using just a description.
The hype associated with artificial intelligence has not lost its intensity. On the contrary, the number of online services continues to grow, and we have already seen almost everything, from radio stations to virtual assistants on WhatsApp. Ethical and legal conflicts are still far from being resolved, but if there is something that can escape that noise, it is the generation of 3D objects.
The people at OpenAI are looking to point in that direction with a new project called Shap-E. In essence, what OpenAI did was publish the code, but a lot of information is still missing. Shap-E's GitHub profile lacks advanced instructions, and doesn't even describe the necessary dependencies, but those who manage to install it will have two advantages: It works offline, and it doesn't depend on OpenAI's API.
Shap-E: The first steps of text to 3D
There are three modes of operation: Text to 3D, image to 3D (that is, a 2D -> 3D conversion), and Encode, which uses Blender to take a 3D object and convert it into something different. Logically, Shap-E needs a lot of firepower, and the first explorers recommend Nvidia hardware to get started. Even with a GeForce RTX 3080, rendering a model can take between four and five minutes.
Its results are definitely basic, and errors appear everywhere… but it's not bad for an initial version. From a (much) simpler installation to performance improvements, we expect Shap-E to gain traction. The potential of text to 3D is gigantic, and if its models ever reach print-quality… the sky is the limit.
Official site: Click here
Source: Tom's Hardware
