Just as generative AI has been stirring things up with 2D images, people are also looking hard at 3D models too. One great little video I stumbled across lists a number of platforms, some of which I had not heard of before. First, the video:
Some of the sites mentioned:
- For search: Asset Ovi (a search engine for 3D models) and ObjVerse (a data set of 800K annotated 3D models).
- Text to 3D models: MasterPiece Studio (commercial product)
- NeRF (2D images to 3D models): Luma AI and 3DFY
Some of the platforms tackle more than one of the problems at a time. Checkout the YouTube video show notes for the complete list in the video.
Note: The video does not cover all available AI 3D model generation products (for example, Common Sense Machines is not listed). But it shows a lot of people are throwing themselves at the problem and making great progress.
The Point-E demo on Hugging Face from OpenAI to me demonstrates that while great strides have been made, there is still a way to go. The following is a photo of our family’s toy poodle.
The following is the Point-E point cloud for “Toy Poodle”.
Amazing likeness, right? No? Maybe if I rotate it?
No? Let me try one more angle…
Not feeling it yet? Well, let’s just say sometimes things in practice don’t always live up to the demos. 😉
My point? AI 3D model generation is being undertaken by many different teams, and some amazing progress is being made. But its not yet a solved problem.