Text to image AI generators: familiar legal questions about this novel, fast-growing technology
What is text to image AI technology? A text to image AI generator is an online tool that creates images based on simple text. To use the technology, simply provide a text prompt (anything ranging from a word or two to an entire phrase or series of sentences), and the artificial intelligence technology takes over, delivering a unique new image based on a training set of images.
Its functions depend on a deep learning algorithm trained to draw on a large dataset of images and their descriptions. Those images could come from anywhere – typically, they’ve been scraped from the Web.
The use of this technology is expanding rapidly. In 2022, we saw a swift rise in the awareness of artificial intelligence generated images. That is because text to image AI generators are relatively easy to use and already are widely available – numerous free generators are available online and the images they generate can be quite impressive.
In September, for example, an image created via the text to image AI generator Midline won first place in the Colorado State Fair’s annual art competition. The winner did experience some backlash from individuals who objected that the winner had not actually created the image - the AI generator did.
Has your image been stitched together from images in the public domain? Or could they be the intellectual property of others?
This all seems simple, but. While the concept itself may seem relatively basic (though the underlying technology is undeniably complex), the process used to source the data, the data being sourced, and the resulting image all generate a number of important legal questions, particularly within the intellectual property landscape.
What intellectual property protection (if any) is available for AI generated images? At a high level, in order to obtain intellectual property protection for an image, design, or logo, you must be the owner of the artwork. Generally, this either means that you developed the intellectual property yourself, on your own, or that you acquired the rights to that IP from the creator. For example, a design firm is hired to create a logo; it assigns all rights in the design to the business that hired it. An artist paints a picture, which is acquired by a museum; the museum then owns the rights to reproduce the painting. But when AI plays a role in the generation of an image, difficult questions arise. Among them: is this a new copyrighted work? If it is composed of IP that is already owned by someone else, then who is able to claim ownership of this new work?
For the time being, The United States Copyright Office seems to be taking the position that purely AI generated images “lack the human authorship” required to support a US copyright claim. This position is currently being challenged in District of Columbia district court by an individual who attempted to file a copyright application for an image generated via artificial intelligence. We will likely see a decision on this challenge in the coming months.
Before using an image generated via text to image technology, a business should think critically because serious questions regarding ownership of AI generated images remain unanswered. Businesses cannot simply assume that such images are free to use without typical clearance considerations. Because you asked the generator to create this image does not necessarily mean that it is legally yours. For instance, has your image been stitched together from images that are in the public domain? Or could they be the intellectual property of others?
These questions are particularly important in the commercial context. To the extent an AI generated image is produced using training sets from materials that are not in the public domain, questions may arise as to whether the resulting image could be an infringement of the source images. Questions beyond copyright could arise as well. In particular, consider whether AI generated art features the image or likeness of an individual (whether famous or not) or includes a well-known character or design. Also consider if the design incorporates an existing trademark (company name, logo design).
All types of rights holders are sure to be looking at AI generated art with a critical eye.
What do next steps look like? When it comes to text to image AI generators, both the technology and the legal considerations relating to it will evolve. Text to image AI software is in its very beginning stages of development and adoption, and we will see swift growth and expansion in the coming years. This includes, but is not limited to, the district court’s decision in the challenge on the US Copyright Office’s position on the copyright protection awarded to AI generated images.
While the IP law issues are not new, the new features of this technology and the resulting images will be a matter of first impression of the courts as they consider the questions prompted by this new technology. We will continue to closely monitor the situation and we will provide further updates on this subject as developments occur.
White House AI Bill of Rights may prompt agency rulemaking and legislation
29 November 2022 .7 minute read
Mark it: 6 big trademark, copyright and advertising trends we are watching for 2023
3 January 2023 .3 minute read