Case Study: Contracting Terms of Use for AI-Generated Art Using DALL-E

  • December 06, 2022
  • Abby Benattar, WeirFoulds LLP

There is a popular saying that goes, “a picture is worth a thousand words.” 

In recent years, a group of engineers at the California-based research lab, OpenAI, have revolutionized this expression with their innovation of DALL-E and its successor, DALLE-2 (“DALL-E”). In November 2022, OpenAI released DALL-E in beta as an application programming interface (“API”), allowing developers to integrate DALL-E into their products. This article describes DALL-E and some core contracting considerations for lawyers advising clients who leverage the technology.

What Is DALL-E?

DALL-E is an artificial intelligence system that is trained to generate, manipulate, and re-arrange images derived from natural language prompts. DALL-E incorporates a dataset of text-image pairs with over 12 billion parameters (i.e., the number of input data that can be transformed into output data) to compose art that is so sophisticated, it can look as though a human created or photographed it (see “DALL-E Demo” example from DALL-E’s website).

The idea of AI-generated art is not new. Much has been written about how Canadian legislators can account for these disruptive technologies.

These questions are still relevant, but the unique capabilities inherent in DALL-E are worth highlighting, not only from a statutory, but also a contracting perspective, especially now that DALL-E has been released in API form.

DALL-E Demo: An output generated by the input, “teddy bears shopping for groceries in ancient Egypt.”

 

 

 

 

 

 

DALL-E Demo: An output generated by the input, “teddy bears shopping for groceries in ancient Egypt.”

Contracting Considerations

If your client leverages DALL-E for commercial purposes, it should have its own set of terms of use with its end-users to reflect the level of liability it takes on in its relationship with OpenAI. As described below, to guide your drafting, you can consult, and where appropriate, flow down some of OpenAI’s key liability provisions found under its recently updated Terms of Use (“Terms”) to the client’s end-users.

Ownership

While OpenAI owns its data, the user is assigned all right, title and interest in and to its outputs. Any art generated by DALL-E will therefore be owned by the person contributing the input that created the art.

Ownership gives the user broader rights than a license does. In the context of AI-generated art, ownership also carries unique risks. For example, several users could contribute the same or similar inputs, and in turn, receive the same or similar outputs. In theory, there could be multiple owners of the same or similar art.

One way to help mitigate the risk of ownership ambiguity is through a general disclaimer under your client’s terms of use. For example, if your client is licensing its DALL-E art to end-users, you can: (a) discuss the merits of disclosing that its images are generated via a third-party API that may reproduce results, and (b) suggest language that explicitly states that all images are provided “as is,” without warranty, including without warranty of non-infringement.

Use Restrictions

OpenAI’s Terms, in fact, explicitly prohibit using DALL-E in a way that causes infringement of third-party rights. Therefore, in addition to the disclaimer, you can pass this restriction down to end-users and discuss with your client whether they wish to mandate a process for their end-users to report any known or suspected infringements. As a business, this would help the client better monitor its fulfillment of OpenAI’s Terms.

Limitation of Liability (“LoL”) and Indemnity

OpenAI disclaims liability for all direct and indirect damages. If a dispute arises, OpenAI caps its liability at the aggregate amounts paid during the 12 months preceding the event giving rise to the claim. This is a standard LoL for online terms of use and can be considered impenetrable for some causes of action.

For your client’s business, it would therefore be wise to strategize passing on a LoL to its end-users that is just as strong as, or even stronger than, OpenAI’s Terms. Your goal should be to flow through these costs so that your client is not out-of-pocket from a claim. This is especially important because under OpenAI’s Terms, your client gives a sweeping indemnity for all claims arising from its use of DALL-E.

Takeaway

DALL-E seems to be an inevitable technology that intends to transform business models and create efficiencies in the creative space. As a result, it is a lawyer’s job to be prepared to manage its risks as they come to the forefront. Since DALL-E is in its beta stage, we can expect it to evolve and for the contracting implications to evolve with it.

51325804.3

Any article or other information or content expressed or made available in this Section is that of the respective author(s) and not of the OBA.