Since their rise to fame, a series of AI tools that generate art from just a few text prompts have sparked a debate: Are these works copyrightable, and if so, who owns them?
With traditional copyright laws centered around human creativity, the ownership and rights of AI-generated works remain a topic of debate. In this article, we’ll explore the copyright for AI-generated art.
Although programmers create AI, their contributions are limited to the design of the algorithm, not the specific work that the AI creates. The artistic choices of AI, such as composition, artistic style, and color palette, are not predetermined but emerge from the complex interplay between algorithms, training data, and user prompts. While prompts do influence the output, they are unlikely to meet copyright standards. Likewise, artists whose work is based on AI cannot predict the final outcome, as the way AI interprets and uses art is beyond their control.
“A Recent Entrance to Paradise” - Stephen Thaler
The absence of a creator in AI has led to many legal controversies. A typical example is the case of Stephen Thaler, whose AI-generated painting “A Recent Entrance to Paradise” was denied copyright by the United States Copyright Office (USCO) because it lacked a human creative element. Thaler’s subsequent appeal to a higher court was also unsuccessful because he explicitly admitted that he did not participate in the AI’s creative process.
Théâtre d’Opéra Spatial
To adapt to the development of new technology, the USCO has issued guidance on the registration of AI-generated works. The guidelines allow copyright registration for human contributions to the work, provided that the AI-generated content is transparently disclosed. However, in related cases such as “Théâtre d’Opéra Spatial” and “Zarya of the Dawn” (visual works created by the AI platform Midjourney), the USCO has refused to protect them because they contain too many AI-generated elements, and it is impossible to separate the creative efforts of humans and machines.
Even though the author of “Théâtre d’Opéra Spatial” used at least 624 different prompts, the USCO still did not recognize his contribution to the creation of the work. The USCO argued that the automated and unpredictable nature of Midjourney makes its creation completely different from the process of human artistic creation, which requires the initiative and control of the author. Unlike artists, who can control every detail of their work, Midjourney users only play a guiding role, not the actual “brains” behind the work.
Contrary to USCO’s reservations, the Beijing Internet Court (BIC) made a groundbreaking ruling in November 2023, granting copyright protection to an AI-generated art with the user considered the author. While acknowledging that the plaintiff did not directly create the final art, the BIC emphasized that the work was created by the AI based on the person’s suggestions.
The court ruled that the plaintiff’s artistic intention and specific choices were single, including the choice of subject (a Japanese idol), specific details (skin color, eye color, hairstyle, pose, etc.), and the iterative refinement through suggestions and parameters, demonstrating intellectual and creative contribution, meeting the requirements of “intellectual achievement” and originality.
However, BIC’s reliance on AI’s lack of legal personality to reject its authorship raises philosophical questions about the nature of creativity: Can a machine, without consciousness or free will, create unique and emotionally charged works? BIC’s approach appears to favor the application of existing copyright law to AI-generated works, even if the law fails to fully reflect the nuances of the creative process.
The future of AI art remains potentially uncertain. AI creates an endless stream of unique creations. The unpredictability and potential for the autonomous creation of AI pose unprecedented challenges to copyright law. While the photographer controls the entire creative process, AI art blurs the lines of authorship, raising questions about who or “what” is the true “creator.”
The future of art and AI is still uncertain, but one thing is certain: the copyright debate will continue to be an important part of this exciting development.
One of the typical disputes is the lawsuit between Getty Images and Stability AI revolving around copyright issues when using data to train AI. Getty Images accused Stability AI of collecting and using a large number of their copyrighted images to develop and train AI models without permission.
According to Getty Images' statement, Stability AI violated copyright by creating new images based on the data warehouse owned by Getty Images, thereby not only causing financial damage but also infringing on Getty's control over its intellectual property.
Stability AI, on the other hand, claims that it only uses images from public sources and does not intentionally violate copyright. According to Stability AI, the input data for its AI models includes images that it believes are taken from public sharing platforms or are already widely used by the public, and as such, its use is legal.
However, the lawsuit has highlighted a complex legal issue: whether using public data to train AI violates copyright. Experts say the lawsuit will set an important precedent for the AI industry in using data from open sources. Some argue that, as AI continues to develop, there needs to be clearer regulations on the right to use public data, especially for copyrighted data such as images or audio. The outcome of the lawsuit is likely to have a significant impact on the way AI companies collect and use data from the Internet. According to the US technology news site The Verge, the outcome of the lawsuit could determine the legal limits of using public data for future technology development purposes.
In that context, some countries have issued stricter regulations on the use of open-source data. In the European Union, the Digital Copyright Law clearly stipulates how organizations and individuals can use copyrighted data to avoid copyright infringement. This regulation aims to protect the rights of content owners when their data is used without permission, especially in the context of AI using data for self-learning and innovation.
More and more creative works created by AI are participating in the co-creation process with humans, raising questions about copyright protection. According to the United States Copyright Office (USCO), works created by AI are not copyrightable, which has been confirmed through recent cases.
However, with the work “Zarya of the Dawn” by Kristina Kashtanova, the USCO took a more flexible stance: The original story created by Kashtanova is copyrightable, but the AI-generated image is not. Kashtanova argued that AI Midjourney is merely a tool, but the USCO disagreed, saying that AI is not as creative as humans and that its output is unpredictable, unlike traditional tools used by artists.
According to the USCO, AI-generated works like those of Midjourney, ChatGPT, or Dall-E are not eligible for copyright protection, registration, or other legal rights in the US. The USCO argued that the images of Midjourney in Kashtanova's work are not original and, therefore, cannot be protected. However, the USCO may consider changing its approach in the future, as AI technology continues to develop and co-creation between humans and AI increases.
The above cases all point to a fundamental legal issue in the current context: whether AI-generated products can be considered copyrighted works. Many countries, such as the US and the UK, still stipulate that only humans have copyright, but AI creativity is blurring this line.
To solve this, these countries are also considering amending the Copyright Law to protect the rights of AI creators or AI owners when AI-generated products become popular. At the same time, the lawsuits of artists such as Sarah Andersen have shown that if the law is not adjusted, it will be difficult to ensure the rights of creators in the AI era.
The USCO has ruled that any art generated by AI models based on text prompts — like Midjourney, Dall-E, or Stable Diffusion — will not be copyrightable in the United States. The decision means the office has equated these prompts to an artist drawing for hire.
US law states that intellectual property can only be copyrighted if it is the product of human creativity, and the USCO office currently only recognizes works created by humans.
So, AI algorithms and machines cannot be authors. The products they create will not be copyrighted.
In addition to images, USCO director Shira Perlmutter also warned that other types of digital art, including poems and books created with tools like Dall-E, Stable Diffusion, Midjourney, or ChatGPT, will not be copyrighted if they are created from human-written suggestions.
“If the traditional elements of copyright in a work are created by a machine, then The United States Copyright Office (USCO) will not register it,” said USCO Shira Perlmutter.
But there is another possibility. The USCO will consider AI-generated content if a human author creates something other than the direct product of a machine. For example, a digital image created by AI from a prompt and then further edited in Photoshop would likely be accepted by the office for copyright registration.
That means an original AI-generated photo would not be copyrightable, but the final product created by the artist could be.
A recent example illustrates this clearly: The USCO Office certified copyright for a graphic novel containing photographs created by Midjourney because the images were generated by an AI model. However, the overall layout, arrangement of images and words within them were created by humans, so it is copyrightable.
AI-generated arts raise complex questions about copyright, ownership, and creative authorship. While traditional copyright laws grant protections to human creators, the status of AI-generated works remains uncertain and varies by jurisdiction.
As AI technology evolves, legal frameworks will need to adapt to address issues of originality, authorship, and fair use. For now, businesses, artists, and AI users should stay informed about copyright laws and best practices to navigate this landscape responsibly.