Shanji FANG
Chinese Attorney-at-Law
Wei Chixue Law Firm
Chinese Attorney-at-Law
Wei Chixue Law Firm
Preface
With the continuous development and popularization of AI creation technology, AI-generated content has emerged extensively in many fields such as literature, art, and news, and the issue of its copyright protection has become increasingly prominent. However, while AI creation brings many innovations and conveniences, it has also triggered a series of disputes. According to the White Paper on Judicial Protection of Digital Copyright released by the Beijing Internet Court in 2024, the proportion of cases involving AI infringement jumped from 3.7% in 2021 to 28.6% in 2023. Due to the particularity of the AI creation process, traditional copyright determination standards are difficult to apply directly, leading to many controversies and uncertainties in practice. This article aims to provide evidence preservation suggestions for right holders when creating works through AI and claiming rights by analyzing relevant legal theories and practical cases, in the hope of offering useful references for solving the difficult problem of burden of proof for copyright of AI-generated content.
I. Criteria for Determining Whether AI-Generated Content Constitutes a Work
Under China’s Copyright Law, a “work” refers to an intellectual creation in the fields of literature, art, and science that has originality and can be reproduced in a tangible form. Among these elements, whether the content falls within the "fields of literature, art, and science" and whether it meets the requirement of “reproducibility” can usually be directly judged based on its external form, and these two elements rarely give rise to disputes in practice. However, when determining whether the content possesses “originality” and reflects the “author’s intellectual creation”—especially in the context of whether AI-generated content constitutes a work—more targeted criteria are often required compared with traditional copyright cases.
1. Possessing Originality
When determining whether a piece of content constitutes a work, “originality” is the core criterion, and this standard does not differ based on whether the content is created by humans or generated by AI. To judge if AI-generated content qualifies as a work under the Copyright Law, the first step is to examine whether it meets the objective requirement of originality. Only if this prerequisite is satisfied is there a need to proceed to subsequent judgments. Second, in the context of AI creation, the assessment of originality is more complex and does not solely focus on the originality of the work itself. Since the AI generation process integrates algorithms, data, and user intervention, multiple factors must be considered comprehensively. While the uniqueness of the generated content itself is important, the intellectual input of the user during the creation process is equally critical. Users guide AI to generate specific content by inputting prompts, setting parameters, etc., and their personalized choices and creative arrangements directly influence the final form of the work and the degree of its originality. For instance, when using AI painting software, a user may carefully design prompts, describe image elements in detail, and repeatedly adjust parameters such as colors and lighting to make the generated image gradually align with their creative concept. The user’s intellectual contribution and aesthetic judgment in this process may enable the final generated painting to have the potential to be recognized as an original work.
Furthermore, it should be noted that if AI-generated content only involves partial adjustments to the color or shape of an original image or video without forming substantial differences from the original work, not only may its originality be denied, but it may also constitute an infringement of the rights to the original work. For example, in Case (2024) Zhe 01 Civil Final No. 10332, the right holder claimed that "the defendant disseminated AI-generated Ultraman images and related LoRA models on an AI platform; when sharing the model, users used the AI-generated Ultraman images as covers and examples; other users could generate images substantially similar to the Ultraman character by inputting prompts and superimposing this model, thereby infringing its right to communicate the work through information networks." After trial, the court held that the AI platform constituted contributory infringement. Although this case primarily focused on the platform’s liability, it also reminds us that even if a user invests creative intellectual effort in the AI generation process, if the generated content lacks substantial differences from the original work, the originality of the result may still fail to be recognized, and there may even be a risk of infringement. Therefore, when right holders claim copyright over AI-generated content, they should also carefully check whether the content constitutes substantial similarity to prior works.
2. Belonging to Intellectual Achievements
In general copyright infringement disputes, the principle of “who claims, who proves” applies to the distribution of the burden of proof. The right holders must prove that they hold rights to the claimed content and that such content meets the requirements for works under copyright law. In traditional creative scenarios, authors typically engage in direct creation using paper and pen, cameras, or conventional image processing software. Therefore, there is usually no need for special proof of whether the author has invested creative intellectual labor; as long as the work possesses originality, the author’s intellectual input is generally presumed to be valid.
However, in the context of AI-generated content, while AI plays a significant role in the generation process, human intellectual input remains the key to determining the nature of the work. AI is essentially a tool that relies on preset algorithms and user instructions for data processing and content generation. It is the user’s intellectual activities during the creative process—such as conception, creativity, selection, and judgment—that constitute the core of determining whether the work possesses originality and legal value. In judicial practice, when courts determine whether AI-generated content is eligible for copyright protection, they usually examine whether the content is a “mechanical generation” or derived from the "human" original expression. If the right holders fail to fully prove their intellectual input, they shall bear the adverse consequences of failing to meet the burden of proof. For example, in a copyright infringement case involving images of cat crystal pendants, the court held that the users should provide original records of the creative process to prove that they selected, modified, and polished the initially generated images by adding prompts, adjusting parameters, etc., thereby demonstrating their personalized choices and substantive contributions to expressive elements such as the composition, perspective, colors, and lines of the images. Since the plaintiff, Feng Mou, failed to provide original records such as flowcharts related to the creative process, his claim of intellectual input lacked evidentiary support and was not recognized by the court.
Therefore, to assert copyright over AI-generated content, it is necessary to further provide evidence proving that the right holder has invested original intellectual achievements. This not only includes integrating specific visual, narrative, or structural elements into the writing of prompts but also means that the user has not only proposed a creative direction but also selected specific forms of expression, endowing the prompts themselves with characteristics of personalized expression. If, after generating the work, the user continues to modify prompts, adjust parameters, or conduct post-processing such as editing, synthesizing, or polishing the generated content, such content is more likely to be recognized as a work within the meaning of copyright law and thus protected by law.
II. Recommendations on Burden of Proof for Copyright of AI-Generated Content
As mentioned above, when the right holder claims rights to AI-generated content, it is particularly critical to explain and prove the elements of originality and investment of intellectual achievements. This is not only because these elements form the basis for a work to be protected by copyright, but also because proving these elements is more complex and challenging in the context of AI creation. For the determination of originality, in addition to whether the AI-generated content itself has “unique expression distinguishable from existing works”, the right holder’s creative ideas, the content of input prompts, and the process of selecting and modifying the generated content can all effectively demonstrate the author’s original intellectual input process for the work. Such evidence can not only help the court rule out the reasonable doubt of “AI independent creation”, but also enable a more accurate determination of whether the work meets the statutory requirements for originality. Therefore, it is particularly important for the right holder to retain evidence of the AI creation process.
However, this does not mean that the right holder can be reassured simply by retaining relevant evidence. If evidence is not retained in a timely manner, or the content of the evidence fails to reflect “creative intellectual input”, there remains a risk that the copyright claim will not be recognized by the court. The same risk exists if the right holder fails to take timely evidence preservation measures or cannot provide evidence to prove creative input. Based on judicial cases, the author puts forward the following specific recommendations regarding the evidence retention issue for right holders claiming rights to AI-generated works:
1. Evidence retention must be timely and complete; ex post facto descriptions are difficult to prove the process of original input.
When using AI for creation, the right holder must establish the awareness of “keeping traces of the process” and timely and accurately preserve every key link and information in the creation process. The right holder shall immediately record each step of the operation during the work creation, including the software version used, the specific instructions input, the modification process, and the generation result of each step. This can not only prove the originality, timeliness, and identity of the right holder of the work, but also provide strong evidence support when disputes arise, serving as an important guarantee for the right holder to safeguard their legitimate rights and interests.
In practice, some right holders do not attach sufficient importance to the retention of evidence for AI-generated content. When claiming copyright, they fail to submit records of the creation process and only attempt to reconstruct the AI creation process through “ex post facto reproduction and description” during court trials. However, this method is usually not recognized by the court in judicial practice, and in most cases, the right holders cannot fully reproduce the original work they claim. For example, in the copyright infringement case involving images of cat crystal pendants, the court clearly stated: "First, the plaintiff failed to submit records of the creation process of the at-issue images in the AI software, and thus cannot demonstrate the specific process by which the plaintiff used this tool to generate the at-issue images. This is crucial because records of the creation process can intuitively show the author’s intellectual input and creative choices when generating the work. Second, the specific results of the at-issue images submitted by the plaintiff under the ‘describe’ command are merely ex post facto descriptions of the at-issue images using the description word generation function in the AI software, rather than a reconstruction of the original prompts or generation instructions. This means that these descriptions cannot illustrate the content of the instructions and prompts input by the plaintiff during the original generation process, and thus cannot reflect his creative ideas and intellectual input. Third, the input situation of the ‘reproduced description’ submitted by the plaintiff cannot objectively reconstruct the original generation process of the at-issue images. From the perspective of the reproduction process, the relevant process is merely an ex post facto simulation conducted by the plaintiff with reference to the at-issue images, which lacks identity and comparability with the original generation process of the at-issue images in terms of hardware and software equipment, network environment, input instructions, and operation steps. It is impossible to infer, based on the aforementioned ex post facto simulation operations, that the plaintiff made corresponding selections, arrangements, and judgments during the original generation process of the at-issue images and devoted creative labor. From the perspective of the reproduction results, there are also certain differences between the ex post facto simulation results and the at-issue images in terms of style, pattern, and composition. Therefore, the evidence in the case is insufficient to confirm that the at-issue images possess originality, and the at-issue images do not constitute works within the meaning of the Copyright Law. The plaintiff’s claims lack factual and legal basis and are not supported by the court.
As can be seen from the aforementioned case, if the right holder fails to timely solidify evidence of the AI creation process and attempts to reconstruct the creation process through “ex post facto descriptions” after a dispute arises, this method is inherently unable to prove the original creation process. This is because the generation mechanism of AI is characterized by significant “generative uncertainty”. The same prompts may yield significantly different results when used at different times, in different hardware and software environments (such as updated AI model versions or differences in device computing power), or even under different network conditions. This is fundamentally different from the characteristic of traditional creation, where “authors can stably reproduce works using the same tools and steps”, making it technically difficult to reconstruct the original work through ex post facto reproduction.
Furthermore, ex post facto reproduction is essentially a form of “backward-looking simulation”. The right holders often adjust the description content based on the existing finished work, causing the “reproduction instructions” they submit to lose the randomness and creativity inherent in the original creation. Such instructions are more like “textual reproductions” of the finished work rather than reflections of the real creative thinking.
From the perspective of evidential weight, such ex post facto evidence lacks “originality” and “relevance”, and cannot form a complete chain of evidence to prove the process of intellectual input during the work’s generation. Naturally, the court will find it difficult to recognize its evidential weight. This further underscores the irreplaceability of real-time traceability and evidence solidification during the creation process. Only by recording operations simultaneously as they occur can the authenticity and integrity of the creation be preserved to the greatest extent.
2. The AI Creation Process Must Reflect the Author’s Original Intellectual Input.
From the perspective of proving a work’s originality, detailed and accurate evidence must not only “fully record the process” but also “highlight intellectual input” — that is, through evidence, it should clearly demonstrate how the right holder, during the AI creation process, guided the final expression of the work through independent thinking and personalized choices. For example, in AI painting creation, retaining records of input prompts, parameter settings, and multiple revisions can intuitively reflect how the right holder guided AI to generate works with unique styles and expressions through distinctive ideas and personalized operations. If AI-generated content is triggered only by simple, general prompts, this essentially amounts to providing a “creative theme” rather than designing the specific expression of the work. Therefore, in practice, such content is generally not recognized as a work protected by copyright law. In a case heard by the People’s Court of Diecai District, Guilin, the court explicitly rejected the copyright claim corresponding to “general prompts”: the at-issue images were automatically generated by AI software, and the user only input general prompts. The user’s prompt — “Title: ‘Heavy Snow’ (one of the 24 Solar Terms); there is a folk saying for this solar term: ‘Preserve vegetables in light snow, preserve meat in heavy snow’; every household cures salted meat” — merely provided a thematic concept. It did not directly or specifically design or create identifiable, original visual elements in the images, such as the specific shape of objects, color matching, spatial layout, lighting effects, and other details that constitute the substantive expression of the images.
In this case, the court further pointed out: “In this case, the user’s input in formulating the prompts is more akin to operating and utilizing the AI software, as well as screening the output results. Although such input may reflect the user’s ideas, its core role lies in activating and guiding the AI’s automatic generation mechanism — rather than personally ‘creating’ or ‘drawing’ the specific expression of the image, or making substantive, identifiable creative judgments, selections, or arrangements regarding the generated content. The user’s act of inputting the aforementioned prompts constitutes an ‘instructive operation,’ not ‘expressive creation’ protected by copyright law. Meanwhile, the AI image generation process is ‘random’: the same prompts produce different images each time, which indicates that the user lacks control over the final expression and cannot effectively demonstrate original intellectual input.”
In sharp contrast, in the case (2023) Jing 0491 Civil First No. 11279, the court upheld the right holder’s copyright claim. The core reason was that the evidence fully demonstrated “original intellectual input.” After trial, the court held that, from the perspective of the generation process of the at-issue images: on the one hand, the plaintiff designed visual elements such as characters and their presentation methods through prompts, and set parameters for image layout and composition, which reflected the plaintiff’s selections and arrangements. On the other hand, after obtaining the first image by inputting prompts and setting relevant parameters, the plaintiff continued to add prompts, modify parameters, and make multiple adjustments and revisions, ultimately obtaining the at-issue images. This process of adjustment and revision also reflected the plaintiff’s aesthetic choices and personal judgments. During the court trial, the plaintiff generated different images by changing individual prompts or individual parameters. It can be seen that when using this model for creation, different individuals can independently input new prompts, set new parameters, and generate different content. Therefore, the at-issue images were not “mechanical intellectual achievements.” In the absence of contrary evidence, it can be determined that the at-issue images were independently completed by the plaintiff, reflected the plaintiff’s personalized expression, and met the requirement of “originality.”
The comparison between the above two cases clearly shows that, to prove the originality of a work, the right holder must demonstrate “substantive intellectual input” during the AI creation process, rather than merely conducting “instructive operations.” Simple prompts cannot enable the user to substantially intervene in the content of the work; the content output by AI is essentially AI’s independent interpretation of a “thematic concept,” and the user cannot claim authorship over such “interpretation results.” After all, different AI systems can output different images based on the same prompts, and even the same AI system can generate countless versions from the same prompts. If copyright attributes are recognized merely because a user inputs a theme, the standard of “originality” will be excessively lowered, and it may even lead to chaos where “multiple users claim copyright for the same theme.” Therefore, when the right holder claims copyright over AI-generated content, the focus of proof should be on demonstrating “creative intellectual input.” For example, the right holder can make detailed modifications and selections to the generated content, integrating personal creativity and aesthetics into the work. In addition, the right holder can also record the thoughts and decisions during the creation process — such as why certain specific instructions were chosen or why certain elements were modified — which can more effectively prove that the originality of the work stems from the right holder’s intellectual input.
3. Adopt Reliable Means of Evidence Preservation
Article 11 of the Provisions of the Supreme People’s Court on Several Issues Concerning the Trial of Cases by Internet Courts clearly stipulates the determination standards for electronic evidence. When preserving evidence, the right holder shall not rely solely on easily tampered methods such as “locally saved screenshots and documents.” Instead, they shall adopt reliable evidence preservation means that meet statutory standards to ensure the authenticity, integrity, and admissibility of the evidence. In practice, blockchain technology and third-party evidence preservation platforms are two relatively mature methods recognized by the judiciary, which can effectively solve the problems of electronic evidence being “easy to tamper with and difficult to trace.”
Specifically, during the creation process, the right holder may regularly upload creation data and operation records to a blockchain or a third-party evidence preservation platform. These platforms can provide timestamps and digital signatures to ensure the integrity and authenticity of the data. When a dispute arises, the right holder can use the evidence provided by these platforms to prove their creation process and intellectual input, thereby better safeguarding their rights and interests. When a creator uses AI software for creation, the platform can record the creator’s operation process in real time — including steps such as inputting instructions, adjusting parameters, and selecting generated results — and store the recorded videos and relevant data in an encrypted manner. During the recording process, the platform will strictly comply with legal provisions to ensure the legality of the recording behavior, and will conduct timestamp authentication on the recorded videos and data to ensure the authenticity and integrity of the evidence.
Third-party evidence preservation platforms also provide professional evidence management and issuance services. When a copyright dispute occurs, the platform can, at the right holder’s request, provide certified evidence documents and relevant supporting materials. These evidence documents and supporting materials have been certified by professional institutions, possessing high legal effect and credibility, and can be recognized and admitted by the court in litigation.
4. Copyright Registration Reduces the Burden of Proof in Rights Protection
The primary function of copyright registration lies in that the registration certificate issued thereby is a form of evidence with official certification effect. In copyright disputes, the registration certificate can directly prove that the right holder enjoys copyright over the work, significantly reducing the right holder’s burden of proof. According to relevant legal provisions and judicial practice, in the absence of contrary evidence, the right holder recorded in the copyright registration certificate is presumed to be the copyright owner of the work. This means that when the right holder presents the copyright registration certificate, the opposing party, if intending to refute the right holder’s copyright claim, must bear the burden of proof to demonstrate that the registration certificate contains errors or is untrue.
Copyright registration also clarifies the creation time of the work and the right holder’s information. Creation time is a key factor in copyright disputes, as it holds great significance for determining issues such as the work’s originality and the existence of prior rights. Through copyright registration, the work’s creation time is clearly recorded in the registration certificate, which provides an accurate timeline basis for the right holder during rights protection.
It should be noted that copyright registration authorities do not review the AI creation process. Therefore, the copyright registration certificate only serves as preliminary evidence to prove copyright. It is still necessary to combine it with evidence detailing "the author’s original intellectual input" to form a complete evidence chain, which can more comprehensively and effectively prove the work’s originality and the right holder’s contribution to the creation.
III. Conclusion
In summary, when right holders create works through AI and claim rights thereto, they should attach great importance to the preservation and solidification of evidence. By promptly recording the creation process, demonstrating original intellectual input, adopting reliable means of evidence solidification, and conducting copyright registration, the right holders can effectively reduce the burden of proof in rights protection and better safeguard their legitimate rights and interests.