Lots of users who wish to create image content from text descriptions are waiting to test OpenAI's Dall-E system. But whether the technology will pass corporate compliance muster could be a gating factor to its ultimate success in the enterprise.
The San Francisco-based AI research lab said on July 20 that it will make the tool available to more users on its 1-million-strong waitlist in the coming weeks. Dall-E, which uses natural language to translate text to images, is already available in beta.
The disclosure came a few months after OpenAI released the second version of Dall-E, Dall-E 2. The research lab said that Dall-E 2 can create more realistic images than the original Dall-E, which was first introduced in 2021 as a 12 billion parameter version of GPT-3.
Enterprise use cases
While many consumers have successfully used Dall-E to create creative memes, or even create music videos, enterprises can also find success with the tool, said R "Ray" Wang, analyst at Constellation Research.
"This is about expediting, automating and using AI to create content," Wang said.
The tool is helpful for enterprises that work in the commercial printing sector or those that might consider creating content for the metaverse.
While there are different ways to build content at scale right now, most of it is manual, Wang said. Dall-E automates that manual process.
Ray WangAnalyst, Constellation Research
In the technology's current iteration, OpenAI will allow Dall-E users to reprint, sell and merchandise their images.
Responsible AI and ethics
In this release, OpenAI takes a responsible AI approach, Wang said.
The research lab said its filters are more effective in preventing users from creating violent, adult or political content. However, OpenAI must show the market how it built the models with the ability to audit the data set and results, said Andy Thurai, another analyst at Constellation Research.
An AI program, written in plain text, can be instructed to edit a picture in a malicious way to post online, which can damage a company's or an individual's reputation, Thurai said. He added that users should exercise care as well as offer governance when they must show a generated image versus an original work by an artist. Also, OpenAI must prove that it can only manipulate G-rated images.
The ethical and private implications of Dall-E might make enterprises skeptical of using the tool, because the tool can misrepresent what is originally said. "Enterprises are less tolerant to mistakes," said Sid Nag, analyst at Gartner.
So even though Dall-E could be a real game changer, compliance regulations might prevent this technology from ever getting in the company door.
OpenAI against the competition
This trial version of Dall-E puts OpenAI a step ahead of competitors such as Google, which released Imagen earlier this year. Wang said vendors are anxious to get these products in users' hands so that they can see how they will use the technology.
"This is really a game of speed," he said. "This is all about building the training data on the back end to see how people are using it [and] what else they can do to modify or add features."
Competitors such as Google will likely watch how the market receives OpenAI before they follow suit, Wang said.
Dall-E users get free monthly credits. Anyone who wants more credits can purchase an additional 115 credits for $15, which allows them to create 460 images.