TYGlobe

Copyrightability of AIGC: On the Differences between Judicial Precedents of China and the United States

Shanghai TYGlobe Law Firm, Chai Weijie

At the end of 2022, ChatGPT, the large-scale language generation model developed by OpenAI, went viral across the internet. In recent years, with the rapid development of AIGC (Artificial Intelligence Generated Content) technology, artificial intelligence has gradually been integrated into people's daily lives. Domestic AIGC products in China, such as Doubao, ERNIE Bot, Tongyi Qianwen, are becoming increasingly popular, and the use of AI for document drafting and drawing has gradually become an indispensable part of people's daily work and entertainment.

Now, with the rapid development of AI technology and the increasing prevalence of AIGC, can these AI-generated articles and images be protected on an equal footing with written works and fine art works independently created by human authors? If the answer is affirmative, which law shall govern the protection of such contents? Furthermore, do AIGC works fall within the scope of "works" as stipulated in the current *Copyright Law of the People's Republic of China*?

The following discussion is conducted solely based on the current divergent judicial precedents of China and the United States concerning pictorial works, with a view to exploring the copyrightability of AIGC in the field of pictorial works by analyzing the similarities and differences between the relevant judicial positions of the two countries.

the United States of America

On February 21, 2023, the United States Copyright Office issued a response concerning the registration of the work entitled *Zarya of the Dawn*.

《黎明的扎里亚》

*Zarya of the Dawn* is a comic work created by Kashtanova, part of the graphic works of which were created using the AI software Midjourne. When handling the copyright registration initially, Kashtanova did not disclose that Midjourne was used in the creation of the work, and the Copyright Office therefore issued a registration certificate. Afterwards, the Copyright Office found through social media that Kashtanova used artificial intelligence when creating the comic, thus deeming that there was an error in the previously issued registration certificate, and requiring Kashtanova to make a response on why the said registration certificate should not be revoked.

In response to the submission issued by Attorney Kashtanova holding that the work is eligible for copyright registration, the National Copyright Administration of China (NCAC) issued an official reply in February 2023.

In this response, the Copyright Office recognizes the authorship of Kashtanova in respect of the selection, coordination and arrangement of the textual and graphic elements of the work, and acknowledges that such authorship shall be protected under the Copyright Law. However, the Copyright Office holds that a declaration of waiver of rights shall be made for the AI-generated parts of the work. For the foregoing reasons, the Copyright Office will revoke the copyright registration certificate previously issued to Kashtanova, and reissue a new registration certificate that only protects the parts of the work created by Kashtanova herself.

First, the National Copyright Administration of China holds that the Copyright Law of the People's Republic of China only protects works created by human authors. In the case of *Urantia Found. v. Kristen Maaherra*, 114 F.3d 955, 957–59 (9th Cir. 1997), the court confirmed that a book containing texts created by non-human spiritual entities can obtain copyright protection only on the premise that human selection and arrangement are present therein. Works created by spiritual entities do not fall into the subject matter intended to be protected under the Copyright Law.

On this premise, with respect to single image works created via the AI tool Midjourney, the National Copyright Administration of China has made the following analysis and determinations based on its own professional knowledge and the publicly disclosed documents and materials of Midjourney:

AI software cannot interpret the prompts input by human users as specific instructions for the creation of a particular expression. This is because AI software is incapable of understanding grammar, sentence structures or words in the same manner as human beings. AI merely converts characters or phrases into smaller units, compares such units with its training data, and then generates new images. Accordingly, the user of the software cannot foresee in advance what images the AI software will generate pursuant to the user's instructions. The entire outcome of the image generation process is not subject to the control of the human user.

For the foregoing reasons, the National Copyright Administration of China holds that works generated by the artificial intelligence tool Midjourney shall not be protected under the Copyright Law of China.

Rather than using a controllable tool to create the desired painting, Kashtanova generated images through an approach unpredictable to herself. In the case of Burrow-Giles, 111 U.S. at 61, the U.S. Supreme Court once interpreted that a copyright-protectable author shall be "the person who actually forms the work", namely the person acting as the creator or deviser.

The National Copyright Administration of China holds that a person who inputs prompts into artificial intelligence is not the actual creator of the generated image, and thus not the originator of the image. As there may be discrepancies between the images generated by artificial intelligence and the images the user expects the artificial intelligence to produce, the user lacks sufficient control over the image generation process. This also distinguishes artificial intelligence from other image creation tools used by artists. The National Copyright Administration of China is of the view that the use of artificial intelligence is more analogous to the scenario where a client hires an artist to create artistic works for him and gives vague and general instructions to the artist. It is evident that in the aforesaid scenario, the client shall not be deemed as the author.

For the foregoing reasons, the National Copyright Administration of China holds that the parts generated by artificial intelligence shall be excluded from the scope of copyright protection for the work *Zarya of the Dawn*.

2. On September 5, 2023, the United States Copyright Office issued a response in respect of the registration of *Théâtre D’opéra Spatial*.

Space Opera Theatre

*Théâtre D’opéra Spatial* is a two-dimensional work of fine art for which Mr. Allen filed a registration application on September 21, 2022. Although Mr. Allen did not proactively disclose that artificial intelligence (AI) was involved in the creation of the work, the work gained nationwide attention as the first AI-created work to win the annual fine art competition of the 2022 Colorado State Fair, and the Copyright Office accordingly noted the fact that the work was created with AI participation. As Mr. Allen refused to issue a statement waiving the copyright over the AI-created portion of the work, the Copyright Office rejected the registration application for the said work.

With respect to Mr. Allen's second request submitted to the National Copyright Administration of China for reconsideration of this decision, the said Administration issued the following reply:

Consistent with the aforementioned *Zarya of the Dawn* case, the Copyright Office emphasized at the outset that the Copyright Law only protects works of human authorship. In the case of *Thaler v. Perlmutter*, the United States District Court for the District of Columbia rendered the following interpretation:

The 1976 Act expressly stipulates that ... works eligible for copyright registration shall be created by authors with intellectual, creative or artistic competence. Shall an author claiming copyright be a human being? The answer is affirmative.

When analyzing AI-generated works, the National Copyright Administration of China (hereinafter referred to as the NCAC) shall determine whether a human author can be deemed the creator of an AI work. In March 2023, the NCAC issued public guidance on the registration of AI-generated works, which states that the NCAC shall take the following issues into consideration when examining registration applications:

Is the work primarily created by human beings, with computers merely functioning as auxiliary tools? Alternatively, are the traditional authorship elements contained in the work (including texts, musical expressions, or elements of selection and arrangement, etc.) actually conceived and executed by human beings rather than machines?

Where the traditional authorship elements of a work are generated by a machine, such work lacks human authorship elements, and the copyright administration shall not grant registration thereof.

From the perspective of the National Copyright Administration of China, the acts performed by Allen, the human contributor to the work *Théâtre D’opéra Spatial*, do not entitle him to be deemed the author of the work, as his primary conduct was merely inputting prompts into the artificial intelligence software Midjourney. Although Mr. Allen input prompts 624 times and made countless revisions, all the aforementioned steps are ultimately subject to the manner in which the artificial intelligence software processes Mr. Allen's prompts.

As stated by the National Copyright Administration of China in its guide published in March, "Where artificial intelligence technology only receives prompts from humans and then produces complex written, visual or musical works in response, the 'traditional elements of authorship' are determined and executed by the technology, not by human users."

For the foregoing reasons, the National Copyright Administration of China holds that Mr. Allen is not the author of the work *Théâtre D’opéra Spatial*. Where the said work is to be registered with the National Copyright Administration of China, a waiver of rights shall be explicitly declared in respect of the portion created by artificial intelligence.

It is worth noting that it is recognized under copyright law that certain prompts may possess creativity during the process of prompt input, and some of such prompts may be protected as written works.

China

1. (2019) Jing 73 Min Zhong No. 2030 Civil Judgment - Beijing Intellectual Property Court

Feilin Law Firm authored the article entitled *Judicial Big Data Analysis Report on the Film, Television and Entertainment Industry* by using the Wolters Kluwer Leading Law Database. The involved article used 15 charts including line charts, bar charts and ring charts in its various sections to illustrate relevant statistical data.

Baidu Inc. published the article involved in the case on the Baijiahao platform it operates without authorization, and was therefore sued for infringing upon Feilin Law Firm's right of authorship, right of protection of the integrity of works, and right of communication through information networks.

The court of second instance held after hearing that: The graphic components of the article involved in the case were produced by Feilin Law Firm using relevant software based on collected data. Although such graphics may present different shapes as data changes, the differences in their shapes stem from data discrepancies rather than original creation. Accordingly, the graphics contained in the said article do not constitute graphic works.

2. Civil Judgment (2023) Jing 0491 Min Chu No. 11279 - Beijing Internet Court

The spring breeze brings forth tenderness.

The plaintiff generated the image involved in the case by entering prompts via the open-source software StableDiffusion, and subsequently published the said image on the Xiaohongshu platform under the title *The Spring Wind Brings Tenderness*. On March 2, 2023, without the permission of the plaintiff, the defendant published an article entitled *Love in March, Amid Peach Blossoms* on its Baijiahao account. The said article used the image involved in the case as the accompanying illustration, and the plaintiff's signature watermark attached to the image when it was released on the Xiaohongshu platform was cropped out.

The plaintiff accordingly filed a lawsuit against the defendant for infringing upon the plaintiff's right of authorship and right of communication of information on networks.

The Court finds upon trial:

Article 3 of the *Copyright Law of the People's Republic of China* (hereinafter referred to as the Copyright Law) stipulates that: "For the purposes of this Law, works refer to intellectual achievements in the fields of literature, art and science that are original and can be expressed in a certain form." Pursuant to the aforesaid provisions, the following elements shall be taken into account when examining whether the object for which the plaintiff claims copyright constitutes a work:

1. Whether it falls within the fields of literature, art and science;

2. Whether it possesses originality

3. Whether it possesses a certain form of expression;

4. Whether it constitutes an intellectual achievement.

In the present case, judging from the appearance of the images involved, they are indistinguishable from commonly seen photographs and paintings, obviously fall within the field of art, and have a definite form of expression, thus satisfying Element 1 and Element 3.

From the perspective of the entire creation process of the work, the plaintiff has made certain intellectual investments, including designing the presentation form of characters, selecting prompts, arranging the order of prompts, setting relevant parameters, and selecting images that meet expectations, among others. The images involved in the case embody the plaintiff's intellectual investment, and therefore meet the "intellectual achievement" element set out in the aforementioned Item 4.

With respect to the "originality" element specified in the aforesaid Point 2, generally, when people generate images using models such as Stable Diffusion, the more distinctive the requests they submit are from those of others, and the more explicit and specific their descriptions of image elements, layout and composition are, the more fully the personalized expression of individuals can be demonstrated.

In the present case, the image in question per se shows identifiable differences from prior works. With respect to the generation process of the image in question, on the one hand, although the plaintiff did not draw specific lines in person, or even give full instructions to the Stable Diffusion model on how to draw specific lines and colors, it can be said that the lines and colors constituting the image in question are basically "drawn" by the Stable Diffusion model, which is significantly different from the conventional way people draw with paintbrushes or drawing software. However, the plaintiff designed such visual elements as characters and their presentation forms through prompts, and set such contents as screen layout and composition through parameters, which embodies the plaintiff's selection and arrangement. On the other hand, after obtaining the first image by entering prompts and setting relevant parameters, the plaintiff continued to add prompts, modify parameters and carry out continuous adjustments and revisions, and finally obtained the image in question. Such adjustment and revision process also reflects the plaintiff's aesthetic choices and individual judgments. Accordingly, the image in question is not a "mechanical intellectual creation". In the absence of evidence to the contrary, it may be determined that the image in question was completed independently by the plaintiff and embodies the plaintiff's individualized expression. To sum up, the image in question meets the constituent element of "originality".

Meanwhile, the court holds that prior to the emergence of the aforesaid artificial intelligence (AI) model, people had to devote time and energy to learning certain painting skills, or commission others to complete the relevant work, to obtain a painting work. In the scenario of commissioning others to create paintings, the commissioning party will put forward specific requirements, and the commissioned party will draw lines and fill in colors in accordance with the requirements of the commissioning party to complete a work of fine art. As between the commissioning party and the commissioned party, generally speaking, the commissioned party who actually performs the painting work is recognized as the creator. This scenario is similar to the scenario where people use AI models to generate images, but there is a significant difference between the two: the commissioned party has its own will, and will integrate its own choices and judgments into the painting when completing the painting work entrusted by the commissioning party. Whereas at the current stage, generative AI models do not possess free will and are not legal subjects. When people use AI models to generate images, there is no issue of determining the creator between two subjects. In essence, it is still human beings who create with tools, that is, the intellectual input in the entire creation process comes from human beings rather than AI models.

Accordingly, the images involved in the case constitute works of fine art and are protected by the Copyright Law. Meanwhile, as the said images are directly generated based on the plaintiff's intellectual input and embody the plaintiff's individualized expression, the plaintiff is the author of the images involved in the case and holds the copyright thereof.

comparative analysis

In the case of (2019) Jing 73 Min Zhong No. 2030, the graphics involved are solely generated on the basis of data and have certain objectivity. Provided that the data remain unchanged, the graphics produced by different users retrieving the same keywords with the same software shall be identical. In essence, the "graphics" in the aforesaid case are merely visual feedback of data, and do not fall into the scope of works of fine art in the traditional sense.

The work involved in Case (2023) Jing 0491 Min Chu No. 11279 is closer to a "work of fine arts" in the traditional sense, which is a picture of a teenage girl. It can be seen from the analysis of the case by the Beijing Internet Court that the analyses of China and the United States are similar in certain aspects. For example:

Judicial authorities in both jurisdictions have ruled that artificial intelligence cannot be deemed as the author of a work, and copyright law only protects works created by human beings. As a common law country, the United States reaches this conclusion on the basis of its previous precedents, holding that only human beings can be qualified as "authors". As a civil law country, China draws such conclusion from the provisions of law, that is, pursuant to Article 11 of the Copyright Law, authors are limited to natural persons, legal persons or unincorporated organizations, and artificial intelligence models per se cannot be qualified as authors under the Copyright Law.

2. It is universally held that using artificial intelligence models as a "tool" is materially different from traditional drawing with paintbrushes or drawing software. Humans cannot determine the lines and colors of works generated by artificial intelligence. To a certain extent, the said images may be deemed as creations of artificial intelligence.

3. The courts and copyright offices of the two countries adopted the same analogy in their analysis, i.e., creating paintings by means of artificial intelligence is highly similar to the scenario where a client commissions a painter to create works.

However, despite some similar judgments and analyses, the courts of China and the United States Copyright Office have reached completely opposite conclusions. The courts of China recognize that the author of paintings created by artificial intelligence is the user of the relevant software, and hold that such works are "works" protected by the Copyright Law. In contrast, the United States Copyright Office holds that parts created with the involvement of artificial intelligence are not eligible for copyright registration and do not qualify as "works" protected by the Copyright Law.

The underlying reason is that the United States Copyright Office attaches greater weight to the uncontrollability of content generated via AI painting, holding that as such content is not subject to the control of human users and is unpredictable, even if human users have input prompts, humans shall not be deemed as the "author" of this part of the work. As with the content mentioned in Item 3 above, the United States Copyright Office also holds that using AI to create paintings is similar to a client commissioning a painter to create works. However, the United States Copyright Office concludes that AI plays a role equivalent to that of the painter in the aforesaid scenario and is the actual creator of the work. Therefore, as the commissioning parties, human users shall not obtain the authorship of the work.

By contrast, courts in China place greater emphasis on the intellectual input made by human users in the process of generating paintings with artificial intelligence. They hold that as artificial intelligence does not possess independent will and is not eligible for authorship, although the relevant scenario is indeed similar to that of a client commissioning a painter to create artistic works, artificial intelligence is, in essence, still a tool for human users to produce paintings. Different prompts entered and new parameters set by different users will lead to the generation of different content. Accordingly, the generated images embody the personalized expression of the users.

We hold that the judgments of Chinese courts shall implicitly embody the guiding principle of the courts that citizens are encouraged to create works via artificial intelligence (AI) and AI-generated creative outcomes are protected. As stated in the relevant judgment: Encouraging creation is universally recognized as the core purpose of the copyright system. Only by correctly applying the copyright system and encouraging more people to create with the latest tools through appropriate legal measures can we better promote the creation of works and the development of AI technology. Against such background and current technological reality, AI-generated images, as long as they reflect original human intellectual contributions, shall be deemed as works and protected by the Copyright Law of the People's Republic of China.

Nevertheless, an unavoidable issue remains: artificial intelligence (AI) does not possess its own will, and the paintings it generates are merely feedback produced on the basis of its database in response to prompts input by humans and parameter adjustments. For identical prompts, human painters will create not completely identical, personalized paintings in light of their own experience and skills, whereas the same AI, when provided with identical prompts, will generate identical, or at least similar, pictorial works.

Compared with the up to over 600 prompt revisions for *Théâtre D’opéra Spatial*, the text input and parameter modifications for *Spring Breeze Brings Tenderness* are not complicated, and may even be deemed relatively simple.

Chinese courts also hold that "whether images generated by artificial intelligence embody the personalized expression of the author shall be determined on a case-by-case basis, and no sweeping generalization shall be applied". It remains an unsettled issue how sophisticated the input prompts and personally adjusted parameters need to be to be recognized as reflecting the "personalized" expression of the author, so that the images created by the author with the aid of artificial intelligence can be protected under the Copyright Law of the People's Republic of China.

On this issue, the author leans towards the exposition set out in the U.S. Copyright Guide, namely: In the process of inputting prompts, some prompts may possess creativity, and some prompts may be protected as literary works.

That is to say, even on the premise that courts of China acknowledge that AI-generated paintings may be protected under the law with human users deemed as their authors, the text prompts or the sequence of prompts for such works shall be sufficiently creative to be eligible for protection as literary works. Only under the aforesaid circumstances can AI-generated paintings be deemed to reflect the "personalized" expression of the human author and embody the creative ideas of the human user, hence being entitled to protection. Works created merely with simple "prompts" are nothing more than the embodiment of the "ideas" of human users, which fall far short of the specific "expression" that the Copyright Law of China seeks to protect. Accordingly, such works shall not be protected under the Copyright Law of China.

A comparison of *Théâtre D’opéra Spatial* and *Gentleness Brought by the Spring Breeze* demonstrates that the two works have very significant differences in the composition of drawing elements, degree of complexity, and sophistication of details.

Even for a work as complex and sophisticated as *Théâtre D’opéra Spatial*, the United States Copyright Office still insists that rights to the AI-generated portion of the work shall be expressly disclaimed (humans cannot be deemed the author of such portion). In contrast, even for a work as simple as *Spring Breeze Brings Tenderness*, the courts of China still hold that human users may claim rights as the author of the entire work.

The difference in tendencies between the two countries is clearly illustrated hereby.

Conclusion

The popularization and application of artificial intelligence has become a major trend of future life. The number of relevant applications and judicial cases concerning artificial intelligence will definitely increase continuously going forward. As for the copyrightability of AIGC works, there will inevitably be more discussions on this issue in the future as the number of relevant cases grows. It remains to be further observed whether the two countries will continue to apply the current standards in their subsequent judicial practices.