DALL-E 2 coming to Microsoft’s Azure AI, by invitation

DALL-E 2 is coming to Microsoft’s Azure OpenAI Service by invitation, allowing select Azure AI customers to generate custom images using text or images. The company made the announcement today at Microsoft Ignite 2022, a conference for developers and IT professionals.

“Mattel is actually already using this for their Hot Wheels cars,” said John Montgomery, corporate vice president for Microsoft’s Azure AI platform. “Designers can actually give it prompts and quickly get ideas and tweak modifications.”

A Microsoft blog post gave an example of Mattel designers typing in a prompt such as “a scale model of a classic car” and DALL-E 2 will generate an image of a toy vintage car, perhaps silver in color and with whitewall tires. Then, the designer could erase the top of the car and type, “Make it a convertible” and DALL-E 2 will update the image of the car as a convertible, and then tweak it to add “pink top.”

In the blog post, Carrie Buse, director of product design at Mattel Future Lab, said she sees artificial intelligence (AI) technology such as DALL-E 2 as a tool to help designers generate more ideas.

“Ultimately, quality is the most important thing,” she noted. “But sometimes quantity can help you find the quality.”

The Azure OpenAI Service is currently available in preview with limited access and has been used by customers and partners to access powerful GPT-3 models for common use cases including writing assistance, natural language-to-code generation and parsing data.

Microsoft says it added a responsible AI layer

Adding DALL-E 2 to Azure OpenAI Service will allow customers to generate creative content backed by Azure’s cloud AI infrastructure, enterprise-grade security and compliance features. Microsoft also claims its built-in responsible AI features will help prevent DALL-E 2 from delivering inappropriate outputs.

The company says it removed images from the model training dataset that contain sexual or violent content. It also maintains Azure OpenAI’s filters remove harmful content from prompts and prevent DALL-E 2 from creating images of celebrities and religious objects, as well as “objects that are commonly used to try to trick the system into generating sexual or violent content.”

On the output side, the Azure AI team added models that remove AI-generated images that appear to contain adult, gore and other types of inappropriate content. 

“We’re taking the model, putting it on Azure and bringing all the enterprise credibility and technologies we have there — the security, the compliance, the regional rollouts, everything else, we’re adding a layer around it, kind of our responsible AI,” said Montgomery. “OpenAI has its layers and then we have additional layers on top.”

Addressing DALL-E ownership issues

The Microsoft blog post also emphasized that the Azure OpenAI Service terms today “does not claim ownership of the output of these services.” Other than for its acceptable use policies, “Microsoft’s terms do not restrict the commercialization of images generated by these services, although customers are ultimately responsible for making their own decisions about the commercial usability of images they generate.”

Those comments come as users and experts continue to raise questions about who owns DALL-E images. When OpenAI announced expanded beta access to DALL-E in July, the company offered paid subscription users full usage rights to reprint, sell and merchandise the images they create with the powerful text-to-image generator. And in late September, OpenAI announced that the research laboratory was removing the waitlist for its DALL-E beta, allowing anyone to sign up — citing improved safety systems and lessons learned from real-world use.

Bradford Newman, who leads the machine learning and AI practice of global law firm Baker McKenzie, in its Palo Alto office, said the answer to the question “Who owns DALL-E images?” is far from clear. And, he emphasized, legal fallout is inevitable. 

“If DALL-E is adopted in the way I think [Open AI] envisions it, there’s going to be a lot of revenue generated by the use of the tool,” he told VentureBeat in August. “And when you have a lot of players in the market and issues at stake, you have a high chance of litigation.”