Skip to content

There are too few nonprofit organizations like CC fighting for the commons – support our vital leadership with an end of year contribution. Donate today!

Zarya of the Dawn: US Copyright Office Affirms Limits on Copyright of AI Outputs

Better Internet, Copyright, Licenses & Tools, Open Creativity, Technology
Close up of the face of Zarya, from the comic Zarya of the Dawn, looking intent in a dark urban landscape with glowing lights.

“Zarya of the Dawn”: detail of an image generated by Midjourney AI from the full graphic novel by Kristina Kashtanova.

In a recent post, we explained why, absent significant and direct human creative input, generative AI outputs should not qualify for copyright protection. We noted that exactly what constitutes enough human input is not entirely clear; while a simple text prompt shouldn’t be enough, other areas will present more complex questions.

Just this week, the United States Copyright Office (USCO) provided a bit more clarity, at least as to its views about US law and works generated by AI. While it affirms some of the limits that we articulated, it leaves open many questions about where exactly the boundaries of copyrightability lie.

This case involves an artist named Kristina Kashtanova who used the Midjourney platform to create their graphic novel, Zarya of the Dawn. and their application for copyright registration over their work. Kashtanova claims that they used “hundreds or thousands” of text prompts and went through “hundreds of iterations” to create their artistic vision. They also note that in some instances, they edited the images that Midjourney produced, and altogether put hours of time and care into what became Zarya of the Dawn.

However, even considering the work that Kashtanova put into producing the graphic novel, the Copyright Office ultimately held that the images produced by Midjourney lacked sufficient human involvement in the creation process to qualify for copyright protection. The USCO held that, even if a human enters text prompts into Midjourney and iterates an image until it becomes what they want, they are still at some level out of control over what the generative AI produces. And for this reason, it is not truly the work of a human author. The Copyright Office wrote that “A person who provides text prompts to Midjourney does not ‘actually form’ the generated images and is not the ‘master mind’ behind them. Instead […] Midjourney begins the image generation process with a field of visual ‘noise,’ which is refined based on tokens created from user prompts that relate to Midjourney’s training database. The information in the prompt may ‘influence’ the generated image, but prompt text does not dictate a specific result.”

As part of its decision, the Copyright Office compared Kashtanova’s use of Midjourney with the work of a photographer and their camera. The agency wrote that, unlike a photographer, users of Midjourney do not have the same degree of control over the final images. Photographers can control many parts of their photographs — the framing, the lighting, the subject, the presentation of the subject, the exposure time, the depth of field, etc — even though the camera ultimately captures the image. Midjourney users, on the other hand, cannot truly control what the AI creates. “The process by which a Midjourney user obtains an ultimate satisfactory image through the tool is not the same as that of a human artist, writer, or photographer […] the initial prompt by a user generates four different images based on Midjourney’s training data. While additional prompts applied to one of these initial images can influence the subsequent images, the process is not controlled by the user because it is not possible to predict what Midjourney will create ahead of time.”

When it comes to a ruling like this, it’s all too easy to think of it in terms of which party won or lost. But that’s too simplistic. For one thing, limitations on copyright can be helpful to creativity, growing the commons by enabling artists to build freely on these works.

What’s more, though Kashtanova lost their registration for the full work, the Office also made clear that there were ways for artists to add creativity such that they do have a copyrightable work incorporating AI-generated content. The Office concluded that Kashtanova’s original text, as well as the selection and arrangement of the text and the images together, are copyrightable. Furthermore, the Office noted that it “will register works that contain otherwise unprotectable material that has been edited, modified, or otherwise revised by a human author,” although it did not have evidence in this case of sufficient changes to reach this threshold. Determining those boundaries will have to be evaluated in future instances.

With that in mind, it’s also important to note that neither the artist involved nor Midjourney viewed this as a ‘loss’ per se. While the artist criticized the decision, they stated it is a “great day for everyone that is creating using Midjourney and other tools…. This covers a lot of uses for people in the AI art community.” Similarly, Midjourney’s general counsel declared that the decision is “a great victory for Kris, Midjourney, and artists” because the Office “clearly [is] saying that if an artist exerts creative control over an image-generating tool like Midjourney” then copyright may apply.

Looking ahead, the exact boundaries of copyrightability for AI-generated works will no doubt continue to evolve. The Office’s statements were based on only one type of tool, Midjourney and its “diffusion” model. Other tools may construct images differently, and some generative AI tools give humans more control over the works that they produce, making it more difficult to determine where in the final output the human creativity ends and the AI begins.

Additionally, the copyrightability of fully autonomously created works is currently being debated in Thaler v. Perlmutter, a case brought against the Copyright Office for refusing to register a work where the claimed author is a machine. That case may offer some guidance on these questions in the relatively near future.

Finally, a Copyright Office decision on registration is not the final determination on the copyrightability of works. Copyright in the United States (and other countries covered by the Berne Convention) does not require registration and the federal courts are not bound by any Copyright Office decision. It is, however, a valuable datapoint on how judges may think about works produced by generative AI when these issues end up before the courts. Certainly this is not the last we will hear about this issue and we will continue watching to see what happens.

Like the rest of the world, CC has been watching generative AI and trying to understand the many complex issues raised by these amazing new tools. We are especially focused on the intersection of copyright law and generative AI. How can CC’s strategy for better sharing support the development of this technology while also respecting the work of human creators? How can we ensure AI operates in a better internet for everyone? We are exploring these issues in a series of blog posts by the CC team and invited guests that look at concerns related to AI inputs (training data), AI outputs (works created by AI tools), and the ways that people use AI. Read our overview on generative AI or see all our posts on AI.

Note: We use “artificial intelligence” and “AI” as shorthand terms for what we know is a complex field of technologies and practices, currently involving machine learning and large language models (LLMs). Using the abbreviation “AI” is handy, but not ideal, because we recognize that AI is not really “artificial” (in that AI is created and used by humans), nor “intelligent” (at least in the way we think of human intelligence).

Posted 27 February 2023

Tags