OpenAI is working on a quick and efficient 3D model-building tool
It has been an incredible year for consumer-facing AI and for OpenAI in particular. The likes of the text-to-image DALL-E 2 tool and the recent internet sensation ChatGPT chatbot have been grabbing headlines all year. ChatGPT in particular marked a key moment as an incredibly powerful AI text generation tool met the infinite creativity of the internet and blew everybody’s minds to the extent that most people think the AI revolution is already well underway and we are living in the world of tomorrow today. In this regard then it could be seen as exciting news that OpenAI has announced details about its latest tool called POINT-E.
POINT-E is able to produce 3D models from text prompts in the same way DALL-E 2 is able to create images. The tool is still very much a work in progress, but it is the progress that amrks the impressive feat here. Normally, creating 3D models in this way requires multiple GPUs working for hours, while POINT-E can do the same task in 1-2 minutes on a single GPU. The research paper outlining the work that is going into POINT-E’s development describes it as falling:
“short of the state-of-the-art in terms of sample quality…”
However, it goes on to say that:
“…it is one to two orders of magnitude faster to sample from, offering a practical trade-off for some use cases.”
It works by first creating a 2D image of the text prompt and then building a 3D point cloud of the image. This is how the tool is able to save so much processing power compared to other methods. Furthermore, a second AI also works on the problem to help lighten the load.
As you can see in the images above, the 3D models it is able to create are still rather rudimentary and very much rely on the 3D point cloud method for pulling the models together. There are clear gaps too in some of the models with, for example, flowers having gaps in their stems or chunks missing from the coffee mug, the traffic cone, and the corgi wearing a Santa hat.
This means that for now, the tool won’t be too useful but as OpenAI continues to refine the technology, it could be useful for a wide array of use cases such as developing plans for 3D printing, designing elements for video games and even movies, and more. If you are interested in checking out the project’s code, you can find it on GitHub.
This stuff has been amazing. The beauty of AI art has surpassed that of traditional art. The group that opposes artificial intelligence in art needs to stop sobbing and realize that it can be used as a tool to produce even more incredible works of art. If they don’t adopt it, they will become obsolete. Even if they succeed in halting AI, it will still exist.