OpenAI Makes a Move for 3D Generative Content with Point-E
By Adam Pease
Earlier this year, OpenAI stunned the internet with the release of its text-to-image AI model, DALL-E 2.
This model made it possible to generate near-photorealistic images with nothing but text input.
Now, OpenAI has released its first open-source text-to-3D model, Point-E.
This blog discusses the news and what it means for the generative content market.
What Is Point-E?
While text-to-image AI has been delivering impressive results for the past year, researchers have struggled to leverage AI to produce convincing 3D models that can be used in actual creative settings.
OpenAI’s approach leverages point clouds—or masses of dots that combine to make a 3D shape—to efficiently assemble objects based on user prompts.
At first glance, Point-E’s results are compelling but limited in their applications.
The point cloud models produced by the AI are low-resolution, primitive 3D objects, not the rich and complex digital creations many have been accustomed to by state-of-the-art manual 3D modeling.
However, despite its low fidelity, Point-E excels in delivering results rapidly.
Its model is capable of generating a 3D object in as little as 1-2 minutes, running on a single GPU, blowing many other competing approaches, which can take hours at a time, out of the water.
When Will Generative Content Come to 3D?
We’ve already seen the turbulence as text-to-image models continue to disrupt traditional content creation industries, such as the recent case of Adobe Stock and the stock image market.
We can expect this trend to continue as the ability to generate digital images grows richer and more precise with time. Is 2023 too soon to expect similar changes to arrive in the market for 3D content?
If Point-E is the only player, the answer is likely no, as OpenAI’s results—though impressive—are a far cry from industry-standard, brand-ready generated content.
And yet, other market players, working in tandem with OpenAI, may bring us closer to that reality.
NVIDIA, for example, has long been engaged in attempts to leverage its leading graphics technology for swift 3D synthesis.
While its results have mostly been kept behind closed doors, there’s a good chance that we may see some of them come to market next year.
At any rate, OpenAI’s choice to release Point-E open-source is a welcome move that will drive the entire market forward.
Generative content is not yet ready to disrupt the 3D industry in the same way it has disrupted the market for conventional images.
Nevertheless, OpenAI’s foray into the market is a step in the right direction.
This blog is a part of the Content AI blog series by Aragon Research’s Analyst, Adam Pease.
Missed the previous installments? Catch up here:
Blog 1: RunwayML Foreshadows the Future of Content Creation
Blog 2: NVIDIA Enters the Text-to-Image Fray
Blog 3: Will OpenAI’s New Chatbot Challenge Legacy Search Engines?
Blog 4: Adobe Stock Accepts Generative Content and Meets Backlash
Have a Comment on this?