Copyright Office Stakes Out Position on Registration of AI-Generated Works

There have been two important developments in recent weeks regarding the U.S. Copyright Office's position on registering works created by the use of artificial intelligence (AI) technology. First, on February 21, 2023, the Copyright Office issued its much-anticipated decision regarding the registration of a graphic novel by artist Kristina Kashtanova that included images generated using the AI tool Midjourney. Then, on March 15, 2023, the Copyright Office issued a policy statement providing its first guidance on the subject of copyright registration for works generated by AI.

This Update will discuss in detail the Copyright Office's rejection of Kashtanova's registration and its subsequent guidance on registration of AI-generated work and provide important takeaways for creators who are using generative AI tools.

Kashtanova Registration Ruling

As noted in a prior Update, the Copyright Office first issued a registration of Kashtanova's graphic novel in September, but then reversed course a month later after it realized that the images were generated using Midjourney. It asked the applicant to provide details of the creative process to show sufficient human authorship to sustain a claim to copyright or the registration would be cancelled. The applicant's response provided great detail about the artist's involvement in the creation of the generated images and argued there was sufficient human input. The response described the artist as engaging in a creative, iterative process that included multiple rounds of composition, selection, arrangement, cropping, and editing of each image. It also argued that the artist's extremely detailed text prompts and other inputs that provided direction to the AI tool were sufficient evidence of human creativity, calling these inputs "the tools by which an author…guides the Midjourney service's generation of images consistent with the author's creative vision."

Despite this detailed response describing the applicant's involvement in creating the images, the Copyright Office denied registration for the images in the novel that were generated using Midjourney because it determined that they were not the product of human authorship. The Copyright Office did allow registration of the graphic novel as a whole, based on the applicant's authorship of the text of the novel as well as of the selection, coordination, and arrangement of the novel's written and visual elements. However, this ruling is considered by many to be a blow for those who advocate for the protectability of AI-generated works.

Copyright Office Finds Lack of Predictability Precludes Human Authorship

The Copyright Office based its finding that there was not sufficient human authorship largely on the unpredictability of the generated output. While it recognized that the artist's prompts may have influenced the generated images, it found that the process was not "controlled" by the artist "because it is not possible to predict what Midjourney will create ahead of time" and "the prompt text does not dictate a specific result." Because of this unpredictability, the Copyright Office concluded the artist was not the "mastermind" behind the images and that it was the Midjourney tool—not Kashtanova—that originated the "traditional elements of authorship" in the images. The Copyright Office distinguished this from the use of editing or other assistive tools, where users take specific steps to control the final image such that it amounts to the artist's "own original mental conception, to which [they] gave visible form."

The Copyright Office did not agree with Kashtanova's arguments that use of detailed text prompts permitted copyright protection of resulting images because the images are the visual representation of "creative, human-authored prompts." Although the Copyright Office acknowledged that prompts themselves may in some cases be sufficiently creative to be copyrightable, it did not find that human input in prompts was sufficient to translate into a copyright interest in the generated work. This result seems to be based on a belief that AI tools act on the ideas the prompts convey (which are not copyrightable) rather than capturing any protected creative expression in the prompts themselves. The Copyright Office equates prompts to a client's providing general direction as to content when it hires an artist to create an image (for which, absent qualifying as a work-made-for-hire arrangement, the Copyright Office views the author as the artist who received the instructions and determined how best to express them).

Edits to Generated Images May Be Protectable

The Copyright Office did recognize the possibility that an artist may make sufficient changes to a generated image such that the edited version of such image is sufficiently creative (and contains a sufficient amount of original human authorship) to be entitled to copyright protection. However, with respect to one of the modified images provided by the artist (which modified the character's lips and mouth), the Copyright Office did not find the changes significant enough to supply the necessary creativity for copyright protection. And with respect to the other change cited by the artist (an image showing an aging face), the Copyright Office indicated it did not have sufficient evidence in the record as to how the image was created to make this determination. Note that where modifications to a generated image would be sufficient to be deemed protectable, such protection would extend only to the modifications that were made.

Reactions

The Copyright Office's focus on the predictability and control of the output to determine whether there was sufficient human involvement, and the seemingly high level of required human input, has been the subject of debate since the decision was issued. In an article responding to the ruling, Kashtanova's lawyer wrote:

"The standard is whether there is a 'modicum of creativity,' not whether Kashtanova could 'predict what Midjourney [would] create ahead of time.' In other words, the Office is incorrectly focusing on the output of the tool rather than the input from the human."

The Copyright Office does not cite specific support for its argument that human authorship requires that the outcome of the generated image be predictable by the human author. Kashtanova's lawyer argues that Jackson Pollock couldn't predict how the paint he used would drip onto the canvas (as he used a process involving random dripping and flicking of paint) and that photographers do not always have control over the subjects of their photographs (claiming that "there are many examples of famous photographs that captured animals, people, or humorous situations entirely by mistake"). He also argues that AI tools are not as "random" or "unpredictable" as the Copyright Office seems to think. He notes that while the exact output may not be predictable, the artist can exert control by using detailed input to design the output to have a specific subject, lighting, content, layout, and feel, and that it should not matter that the subject, lighting, content, and layout are generated through prompts instead of captured with a device, such as a camera.

It remains to be seen whether the Copyright Office would view human input that serves to influence the output, but not fully control it, as ever sufficient to meet the human authorship requirement. The Copyright Office appears to be indicating that the answer is no, but it seems likely that this decision will be appealed, and a court may have a different take on the questions presented in this case. Kashtanova's lawyer points out that the standard for human creativity required for copyright protection is fairly low ("a modicum of creativity"), which arguably supports the position that even a minimal amount of creative human input or intervention should be sufficient. Where the level of input by a human artist (whether through detailed prompts, other inputs, or otherwise) has a significant creative influence on the generated image (e.g., by providing direction on things like composition, lighting, subject matter, and mood, that were recognized by the U.S. Supreme Court in the Burrow-Giles case as sufficient to show human authorship), it is possible that a court could view such input as providing sufficient human authorship. However, this would be a significant departure from the Copyright Office's view of the human authorship requirement.

Thaler Case

Note that this case is quite a bit different than the Thaler case that is currently in front of a federal district court in the U.S. District Court for the Eastern District of Virginia. While that case also stems from a rejection of registration by the Copyright Office of artwork that was created using an AI program, Dr. Thaler does not attempt to argue that there was sufficient human involvement in the artwork. Rather, he is challenging the human authorship requirement itself and seeking to have the "Creativity Machine," an AI system he developed, recognized as the author of a work.

Copyright Office Guidance on Registration of AI-Generated Works

Shortly after issuing its ruling on the Kashtanova registration, the Copyright Office issued a policy statement regarding its registration practices for registering works created using generative AI systems. The statement does not provide much new information and largely mirrors what the Copyright Office has already stated in the Compendium of U.S. Copyright Office Practices (the Compendium) and in its recent decisions not to register the works created by Thaler and Kashtanova (discussed above). First, the Copyright Office reaffirmed that human authorship is a threshold issue for copyrightability in its view. Second, it reiterates the statement from the Compendium that characterizes the crucial question as "whether the 'work' is basically one of human authorship, with the computer [or other device] merely being an assisting instrument, or whether the traditional elements of authorship in the work (literary, artistic, or musical expression or elements of selection, arrangement, etc.) were actually conceived and executed not by man but by a machine." Although it notes that the answer will depend on how the tool operates and how it was used to create the final work, it makes a clear statement that works generated by AI systems solely in response to a user prompt fall into that category of uncopyrightable, machine-authored works, even if the prompts are sufficiently detailed to themselves be copyrightable.

Left unaddressed in the Copyright Office's policy statement is under what circumstances, if any, there could be sufficient human input such that AI-generated content could ever be found to meet the Copyright Office standard for sufficient human authorship. The Copyright Office did not provide any examples of what degree of control the user of a generative AI system would need to exercise over the formation of a work to be regarded as the work's author. Rather, the Copyright Office only seems to recognize the potential for a copyright interest in any larger work with embedded generative elements (where the selection and arrangement that is exercised over generative elements is sufficiently creative) or in modifications made to material originally generated by AI technology that are sufficiently creative to meet the standard for copyright protection.

Guidance for Applicants

The Copyright Office statement concludes by providing specific guidance for copyright applicants on how to address the inclusion of AI-generated content in a work submitted for registration (which it notes that applicants have a duty to disclose). The guidance states that:

• Individuals who use AI technology in creating a work and wish to claim copyright protection for their contributions to that work should provide a brief statement in the "Author Created" field that describes the authorship contributed by a human.
• Applicants should not list an AI technology or the company that provided it as an author or co-author simply because the applicant used it when creating their work.
• AI-generated content that is more than de minimis should be explicitly excluded from the application by providing a brief description in the "Limitations of Claims" section. Applicants who are unsure how to proceed can simply provide a general statement that a work contains AI-generated material, and the Copyright Office will contact them.
• Applicants who have already submitted an application for a work containing AI-generated material should make sure they disclosed that material, and if not, they should take steps to correct the information.
• Applications that have already been processed and resulted in registration need to be corrected in the public record by submitting a supplementary registration.
• Applicants who fail to properly disclose information about AI-generated material in their applications or to update the public record after obtaining a registration for material generated by AI risk losing the benefits of the registration.

New Copyright Office AI Initiative

At the same time it issued its policy statement, the Copyright Office also announced a new initiative to examine the copyright law and policy issues raised by AI. This initiative will address both the scope of copyright in AI-generated work as well as the use of copyrighted material to train AI models. The Copyright Office is hosting several public listening sessions this spring, which will provide opportunities for participants to discuss issues related to the use and impact of generative AI in creative fields. And later this year, the Copyright Office plans to publish a notice of inquiry to solicit public comments on copyright issues arising from the use of AI.

Takeaways

• Anyone using generative AI tools needs to be aware that there is great uncertainty as to whether they will be able to claim copyright protection in the output, regardless of the creative process engaged in. Using the recent decisions as a benchmark, it seems the Copyright Office will almost certainly refuse registration of an unmodified computer-generated image.
• Nonetheless, this is not likely to be the last word on this subject, and to maximize potential arguments for human authorship, it will still be important to make sure (and to document) that there is significant human input in the image generation process (whether through detailed prompts or otherwise) that go beyond mere ideas and that can be shown to significantly shape and control the final creative output.
• It is also important to clearly document any changes made by a person to generated images, as the Copyright Office did indicate it would register works that contain otherwise unprotectable material that have been edited, modified, or otherwise revised by a human author if the new work contains a "sufficient amount of original authorship" to itself qualify for copyright protection.
• Applicants who are attempting to register, or who have already registered, works that include more than a de minimis amount of AI-generated materials must adequately disclose the inclusion of such materials or risk losing the benefits of registration. Such applicants should review the Copyright Office guidance (described above) when preparing applications (or to determine whether they need to update their application or registration).

© 2023 Perkins Coie LLP