AI-Generated Art and Its Implications for Copyright Law: Who Owns Creativity?

Authors

Ophelia Lee

Introduction

Artificial intelligence (AI) has reshaped countless domains of human activity, but its profound impact on the arts has raised urgent questions about the meaning of creativity, the boundaries of authorship, and the future of intellectual property protection. Whereas traditional art forms—painting, sculpture, photography—often rely on unmistakable human inputs, AI-generated art emerges from processes wherein an algorithm autonomously produces unique creative outputs based on learned patterns in massive datasets (Manovich, 2019). These outputs can rival or surpass human-generated artwork in terms of sophistication, detail, and visual impact, fueling debates in artistic, philosophical, and legal circles.

Within the legal context, copyright law faces novel challenges: it has long presupposed that a human creator originates protectable works. Indeed, at the very core of copyright jurisprudence is the requirement that an “author” must inject some measure of human originality into the work (Goldstein, 2017). As AI systems grow increasingly complex—particularly generative models such as Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and transformer-based diffusion frameworks (Goodfellow et al., 2014; Ramesh et al., 2021)—they now exhibit an unprecedented capacity for autonomous creation, sometimes producing outputs that seemingly lack a direct human spark.

In this context, one can ask: Who, if anyone, owns the copyright to AI-generated works of art? Or more specifically, does the user who types textual prompts into a creative AI platform hold the legal rights to the resulting image? Are these artistic outputs co-owned by the user and the developer? Or do they remain in the public domain under the rationale that only human authorship qualifies for legal protection (U.S. Copyright Office, 2019)? Without question, this is a broad topic that overlaps with technology, jurisprudence, philosophy, and economics. However, the present research paper narrows its inquiry to a particular subset of the broader debate: it focuses on the role of user inputs (prompts, iterative refinements, and user-directed feedback) in shaping AI-generated art, and whether that user involvement can confer the legal status of authorship necessary to claim copyright.

Although the question of AI “authorship” has been variously explored, from radical proposals that computers themselves might hold intellectual property rights (Samuelson, 1986) to arguments insisting that any autonomy on the machine’s part eliminates protectability (Rosen, 2022), far less attention has been paid to the middle ground: the significance of the user’s textual or conceptual directions in guiding the generative process. Modern platforms—DALL·E, Midjourney, Stable Diffusion, NightCafe—often revolve around “prompt engineering,” whereby users carefully craft textual commands specifying style, content, and mood (Rombach et al., 2022). Some users engage in multiple iteration cycles, adjusting parameters, selecting favored outputs, and discarding less desirable results. This user-driven curation might be understood as a form of active co-creation, akin to a director guiding an improvisational performer.

Yet the law has not kept pace with this new dynamic, and courts face pressing questions. If the user’s prompts are short or vague, perhaps it is the machine’s creative function that dominates. If the prompts and iterative refinements reflect a high level of human originality, one can argue that the user meets the threshold of “authorship” recognized by copyright principles (Gervais, 2020). This paper investigates those possibilities, beginning with an overview of the AI technologies that generate art. It then examines traditional copyright fundamentals, focusing on requirements of human authorship and minimal creativity. From there, the discussion turns to how user prompts and iterative steps could satisfy or fail that standard. Multiple legal theories—ranging from the “tool” analogy to joint ownership frameworks—are considered, followed by an exploration of relevant corporate policies and the limited but telling real-world disputes to date. Ultimately, while acknowledging the unsettled nature of the debate, we propose that user input can sometimes rise to the level of authorship but that it remains heavily context-dependent, requiring nuanced legal evaluations.

Overview of AI-Generated Art

Key Technological Approaches

To understand the copyright complications at stake, it is essential to appreciate how contemporary AI algorithms produce novel visual works. Over the last decade, AI-driven creative pipelines have evolved from rudimentary pattern-matching to sophisticated generative models capable of remarkable diversity, style transfer, and realism (Galanter, 2016).

Generative Adversarial Networks (GANs)

Introduced by Goodfellow et al. (2014), a GAN comprises two neural networks—a generator and a discriminator—pitted against each other. The generator attempts to produce outputs (images, for instance) that mimic real data from a training set, while the discriminator evaluates whether each image is genuine or generated. Through iterative training, the generator refines its ability to create convincingly “real” imagery. Variants of GANs, such as StyleGAN (Karras, Laine, & Aila, 2019), have yielded synthetic portraits, landscapes, or abstract pieces that rival actual photographs or established artistic styles (Elgammal et al., 2017).

Variational Autoencoders (VAEs)

VAEs learn a latent representation of data. An encoder network compresses inputs into a probabilistic latent space, while a decoder reconstructs that data (Kingma & Welling, 2014). By sampling from the learned latent distribution, VAEs can generate new outputs that incorporate the statistical patterns of the training data. Although VAEs are often overshadowed by the more flamboyant successes of GANs, they remain essential for tasks like disentangling attributes of images and allowing more stable generation processes (Galanter, 2016).

Transformer-Based and Diffusion Models

More recent advances in AI art rely on large-scale transformer architectures originally popularized in natural language processing (Radford et al., 2021). Systems like DALL·E or Stable Diffusion combine text prompts with image generation, effectively bridging language understanding and visual synthesis (Ramesh et al., 2021; Rombach et al., 2022). They can integrate style references (e.g., “in the style of Van Gogh”), context-specific content (“a futuristic cityscape with flying cars”), or complex aesthetic instructions, enabling users without specialized coding skills to produce high-quality AI art in seconds. In many respects, these diffusion-based or transformer-based approaches have ushered in a new era of mainstream AI creativity, democratizing access to generative art tools.

Evolution of AI Art: From Experiments to Mainstream Adoption

Historically, computational creativity could be traced back to artists and researchers experimenting with rule-based algorithms, fractal generation, and evolutionary art (McCormack, 2007). However, these older methods generally required direct programming knowledge and had limited realism or expressiveness. The mid-2010s saw a revolution, with deep learning facilitating leaps in style transfer and generative art. Projects like Google’s “DeepDream” (Mordvintsev, Olah, & Tyka, 2015) captured public imagination by revealing the “dreamlike” patterns inside neural networks. GAN-based exhibitions followed, culminating in high-profile auctions of AI-generated works (Barrat, 2018).

Today, AI art has reached a point where hobbyists and professionals alike can harness it for a range of creative tasks: designing concept art for video games, building unique brand logos, prototyping architectural designs, or even drafting entire art exhibitions in a fraction of the time once required (Thornburg, 2021). Many see these tools as augmenting human creativity, but this augmentation challenges existing legal norms about the locus of artistic authorship. For instance, a conventional painter is undeniably the “author” of their canvas, but if a user inputs text into DALL·E and obtains a final image generated through advanced neural computations, the attribution of authorship becomes far murkier (Manovich, 2019).

Copyright Fundamentals and Human Authorship

Copyright law across most jurisdictions generally revolves around three foundational elements:

Originality: The work must originate from the author rather than be copied from another source, and it must embody at least a minimal spark of creativity.

Authorship: Traditional copyright theory supposes a human being undertakes the creative act (Goldstein, 2017).

Fixation: In U.S. law and many other systems, the work must be recorded or embodied in a sufficiently stable form.

Given that this legal structure was conceived when creative works were typically handcrafted or manually authored, it is unsurprising that AI-generation challenges numerous assumptions (U.S. Copyright Office, 2019). Yet understanding these doctrines remains vital to discerning how they might apply (or fail to apply) to users guiding AI processes.

Originality Requirement

In the United States, Feist Publications, Inc. v. Rural Telephone Service Co. (1991) established that even a minimal degree of creativity is enough to confer originality. Subsequent cases indicate that the creative choices need not be monumental but must be discernible and attributable to a human mind. If a user’s textual prompt or series of prompts is simplistic—e.g., “generate a black-and-white cat portrait”—it may be argued that the user’s contribution lacks the required creative input, effectively delegating the entire aesthetic process to the AI. However, if the user’s prompt is highly detailed—perhaps specifying stylistic influences, composition details, lighting preferences, color palettes—and the user iterates multiple times, a stronger claim emerges that the user engaged in creative decision-making (Gervais, 2020; Rosen, 2022).

Human Authorship

Copyright laws globally have remained steadfast in requiring human authorship. The U.S. Copyright Office offers explicit guidance: if a work is created entirely by non-human means, that work is ineligible for copyright protection (U.S. Copyright Office, 2019). A famous example is the so-called “monkey selfie” case, where a macaque pressed the shutter button. The courts held that a non-human animal cannot be an author (Naruto v. Slater, 2018). By analogy, if an AI truly generates the core expressive elements independently of human involvement, it might be barred from copyright. Some commentators question whether this standard is appropriate in an age of creative machines (Bird & Mendis, 2020).

Critically, though, an AI is not a living creature. It is a software-based system that can be conceptualized either as an autonomous agent or as a sophisticated tool. If the latter, we might interpret the user as the ultimate “author” who uses the AI as an instrument, akin to how a photographer uses a camera (Burrow-Giles Lithographic Co. v. Sarony, 1884). Yet such an analogy is imperfect because a camera mechanically captures reality as pointed by the photographer, while an advanced AI might generate brand-new content that cannot be fully predicted or controlled (Mendis, 2020). Thus, whether the AI is functionally subordinate to human directives remains a case-specific question.

Fixation

In practice, AI-generated artworks are almost always “fixed”: once the system outputs an image or a video, that file can be saved. Although ephemeral or dynamic works (e.g., real-time generative installations) might complicate this requirement, for still images and typical artistic outputs, the act of generating and downloading them from an AI platform usually satisfies fixation (Carroll & Zheng, 2021). This dimension, then, is rarely a stumbling block compared to the bigger hurdles of originality and authorship.

The Role of the User in AI-Generated Art

Prompt Engineering as Creative Control

Modern AI art systems frequently rely on users’ textual or parameter-based inputs to guide the generation process. This phenomenon—commonly referred to as “prompt engineering”—empowers users to shape outcomes by specifying themes, references, or aesthetic nuances (Rosen, 2022). In some sense, the user’s words become the “blueprint” from which the AI draws, even though the final expression is heavily influenced by the AI’s training data and learned latent representations (Rombach et al., 2022).

Consider a scenario:

A user instructs the AI: “Create a panoramic view of a misty forest at sunrise in the style of Japanese ukiyo-e woodblock prints, with elongated tree silhouettes and subtle pinkish-hued fog.”

The AI responds with an image that incorporates those stylistic cues.

The user then refines or iterates, perhaps specifying, “Increase the emphasis on the horizon. Add a small figure in the distance wearing a red kimono. Maintain the woodblock aesthetic but introduce a slightly more modern color palette.”

In each iteration, the user’s creative intention is layered onto the machine’s generative capacity. This iterative loop can extend over multiple steps, enabling the user to reject certain outputs, highlight favored aspects, or tweak the instructions (Thornburg, 2021). The question becomes: does this curation and user-driven direction suffice to make the user the “author”? Or is the user merely offering suggestions while the AI’s internal intelligence is generating the “heart” of the expression?

Some argue that if the user’s instructions define the essential compositional elements—style, content, color scheme—and the user meaningfully curates or selects from the AI’s output, that user might be performing the same essential function as a director or orchestrator of creative choices (Leval, 2022). Others respond that the AI’s training on billions of images effectively overshadows any unique input from the user, reducing the user’s role to that of a “prompt pusher” who does not originate the ultimate arrangement or distinctive artistic features (Rosen, 2022). This tension animates the core legal puzzle.

The Spectrum of User Involvement One way to approach the question of user authorship is to conceptualize a spectrum of involvement:

Minimal Input The user types a basic prompt (“generate a cat portrait”) or even clicks a random button with no further curation. The user invests negligible creative insight.

Moderate Input The user’s prompt is moderately descriptive, specifying color palettes or referencing artistic styles, but does not engage in further refinements.

Deep Prompt Engineering The user writes detailed prompts incorporating multiple style references, compositional guidelines, and emotional tone. The output is heavily shaped by these instructions.

Iterative Co-Creation The user systematically refines prompts across several steps, discards undesirable results, merges partial successes, and invests significant conceptual labor in the final composition.

Hybrid or Mixed Media The user merges AI outputs with hand-drawn or otherwise manually created elements, combining them into a final piece that blends machine generation and human artistry.

At the lower end of this spectrum, claims of copyrightable authorship by the user are likely weak because the AI is doing most of the “heavy lifting” with minimal creative direction from the user (U.S. Copyright Office, 2019). However, as we move further along, where the user invests substantial mental effort, specificity, and editorial control in shaping the final product, a stronger argument arises that the user meets the minimal creativity threshold. The law has yet to codify such a spectrum, but logically it captures the varying degrees of user involvement encountered in real-world AI art creation (Gervais, 2020).

Copyright Theories for User Involvement

1. The Tool Theory (Human-Directed Creation)

Under this framework, the AI is merely a tool guided by a human author who retains creative control. Courts have analogized cameras, typewriters, or even software programs as tools in the hands of their operators (Burrow-Giles Lithographic Co. v. Sarony, 1884). Proponents of this view argue that so long as the user exerts genuine artistic control—deciding on the subject, composition, and style—the user’s intangible mental concept is realized through the AI, thus making the user the rightful author (Bird & Mendis, 2020).

Pros: Aligns with the long-standing legal acceptance of technological aids. Encourages innovation by clarifying that people who use AI remain owners if they truly direct the creative process.

Cons: Strains credulity when the AI exhibits emergent or unforeseeable features, making it unclear whether the human actually “controlled” the final expression (Gervais, 2020).

2. Joint Authorship Between User and Developer

Another approach posits that both the user and the software developer might hold joint authorship if the developer’s creative contributions (e.g., designing the model’s architecture, curating training data, tweaking hyperparameters) are integral to the final expression. The user’s prompts combine with these developer choices to yield the image (Collins, 2019). The impetus here is that developer decisions can be highly creative, shaping the aesthetic “taste” of the AI. For instance, a developer might embed biases toward certain color palettes or styles, effectively co-creating the resulting artworks.

Pros: Acknowledges that software developers do more than just provide a blank, neutral tool. In advanced AI systems, the developer’s design choices can strongly influence outcomes.

Cons: Imposes burdensome complexity. The developer typically does not intend to co-author each user-generated piece. Moreover, EULAs may disclaim developer ownership or assign rights to the user.

3. AI as Author with No Copyright

A radical position is that if the AI is the true composer of the work, no copyright attaches (U.S. Copyright Office, 2019). This parallels the “monkey selfie” logic—non-human authors cannot hold copyright, and no one else can claim authorship if the creation is essentially autonomous. Such works might therefore reside in the public domain, available for all to use freely.

Pros: Consistent with established doctrine. Prevents monopolization of works that arguably lack direct human creativity. Potentially fosters open innovation in AI-driven creativity.

Cons: Users and investors in AI might lose incentive to produce advanced systems if no legal protection is possible. Also, it can lead to confusion in scenarios where human input is significant but overshadowed by the AI’s autonomy (Samuelson, 1986).

4. Copyright in the User’s Unique Contributions Only

An intermediate proposition holds that only the user’s explicitly traceable, human-authored components are protected (Mendis, 2020). If the user merges AI-created backgrounds with the user’s own sketches, then that integration is protectable, but the AI-generated portion remains unprotected. Alternatively, the user might add textual elements or modifications after generation, and those changes could be copyrighted if they meet originality standards.

Pros: Fine-tunes the scope of protection to purely human-made aspects. Preserves established norms around public domain expansions.

Cons: Hard to delineate precisely which aspects originated from the user and which from the AI. Could lead to complicated disputes over the boundary between user creativity and AI spontaneity.

Real-World Case Studies and Policies

Corporate TOS and Contractual Approaches

In the absence of a definitive legal framework, many AI art platforms attempt to clarify ownership issues via Terms of Service (TOS). Some early versions of TOS for generative art tools gave the platform broad rights, reserving the ability to use or redistribute user-generated outputs. Over time, user backlash and competitive pressures have prompted updates:

Midjourney: Initially claimed wide licenses to user-created imagery, but later updated terms to clarify that paying users could hold rights to their outputs, albeit with certain usage caveats.

OpenAI (DALL·E): At first, the company was ambivalent about letting users claim ownership, but more recent policies explicitly grant users the rights to images they generate, though retaining a license for the company to analyze or display them under certain conditions (OpenAI, 2022).

Such contractual solutions can momentarily quell disputes by ensuring that users have some sense of control. However, these TOS do not override statutory requirements—if the law deems AI outputs unprotectable, a TOS cannot conjure valid copyright from thin air (Thornburg, 2021). Still, in practice, these private agreements often reduce litigation risk since both parties abide by the negotiated ownership terms.

Litigation and Registration Attempts

While large-scale legal battles remain relatively rare, some conflicts shed light on how regulators might treat AI art. In multiple documented attempts, applicants have submitted AI-generated artworks for copyright registration, only to see rejections or disclaimers from the U.S. Copyright Office (U.S. Copyright Office, 2019). In one instance, the applicant openly stated that no human contributed to the final piece beyond minimal prompting. The Office refused registration on the grounds that the submission did not satisfy the requirement of human authorship.

Other controversies revolve around claims of unlawful copying: some artists allege that AI systems, trained on billions of images, may produce outputs that too closely mirror existing works or replicate an artist’s distinctive style (Anderson, 2022). These concerns raise secondary copyright issues: if an AI’s training process effectively ingests copyrighted images, are the outputs “transformative,” or might they constitute unauthorized derivatives? Courts have not yet provided definitive rulings, but the scenario underscores the complexities in assigning liability when an AI’s “inspiration” emerges from a massive, unstructured dataset.

International Variations

Different legal systems exhibit distinct nuances:

United Kingdom: The Copyright, Designs and Patents Act 1988 includes a clause for computer-generated works, stating that if there is no human author, the “person by whom the arrangements necessary for the creation of the work are undertaken” is considered the author (CDPA, 1988, sec. 9(3)). Some interpret this as granting a form of copyright to whoever operates or orchestrates the computer system, potentially supporting user authorship claims if they provide the essential prompts or instructions (Mendis, 2020).

European Union: The EU typically requires “the author’s own intellectual creation,” emphasizing the necessity of human intervention. The overarching principle suggests that purely AI-driven works do not meet the standard, though formal pronouncements remain sparse (European Commission, 2020).

Asia-Pacific: Jurisdictions like Japan and Singapore have begun exploring specific AI-related IP guidelines, but concrete case law is minimal. Japan, for instance, has considered adopting broad “data creation rights” to incentivize AI-based industries (Bird & Mendis, 2020).

The net result is a patchwork environment, where the outcome of an AI authorship claim could vary significantly based on local statutes or judicial interpretations.

Narrowing the Inquiry: When Do User Prompts Amount to Authorship?

As discussed, the controversies swirl around whether the user’s textual or conceptual inputs can truly constitute the creative spark that copyright law seeks to protect. Within this narrower scope, we might identify three major factors that influence the legal analysis:

Detail and Originality of the Prompt: If the user’s prompt is highly elaborate, specifying unique or personal elements (e.g., referencing a user’s private experiences, personal sketches, or unusual conceptual mashups), it arguably contains the user’s creative authorship. The more generic or trivial the prompt, the less likely a finding of authorship.

Control Over the Generation Process: How extensively does the user direct or supervise the AI? Do they accept the first output uncritically, or do they engage in multiple feedback loops, effectively “sculpting” the final work from the AI’s successive drafts? Repeated curation demonstrates creative judgment akin to an art director guiding an illustrator.

Human Comprehension and Intention: Courts often look for evidence of a deliberate,  human-driven concept. If the user can articulate a prior mental image or intention that the AI was merely executing, that can bolster claims of authorship (Leval, 2022). Conversely, if the user’s involvement is random or so minimal that they cannot claim any conceptual design, authorship is weak.

Threshold of Creativity: A Low Bar?

It is well established that copyright does not demand high artistry—only a minimal level of creativity (Feist, 1991). On that ground, even short textual expressions have been found copyrightable if they reflect some unique arrangement of words or imaginative flair (Goldstein, 2017). By analogy, certain prompts might themselves be protectable textual works, and the question shifts to whether the resulting image is a “derivative work” of that textual prompt or an entirely new creation. However, analyzing the generative process is more complex than typical derivative works, since the AI draws on an enormous training corpus, not solely on the user’s prompt (Rosen, 2022).

One possible outcome is that the user owns a copyright in the prompt text itself—especially if it is original writing—but not automatically in the AI’s resulting image unless the user’s creative direction can be demonstrated in that image’s final arrangement, composition, or style. This could lead to arcane distinctions where the text is clearly protected, but the final output remains unprotected or partially protected. Such complexities highlight the difficulty in fitting AI generation into established copyright categories.

Emerging Policy Proposals and Potential Resolutions

Given the legal uncertainty, policymakers, scholars, and industry stakeholders have floated various solutions. Some propose targeted statutory reforms or clarifications in administrative guidelines, while others believe the existing framework suffices, provided courts interpret it in a nuanced manner.

Legislative Clarity

One straightforward approach is for legislative bodies to enact explicit statutes regarding AI-authored works. Such legislation might:

Require meaningful human oversight as a condition for copyright. The law could state that if the user cannot demonstrate a measurable degree of creative control, no copyright subsists.

Introduce limited-term protection for AI-assisted works, balancing the desire to encourage AI innovation with the impetus to avoid indefinite monopolies on machine-generated culture (Gervais, 2020).

Define the threshold of “active guidance” or “prompt engineering” needed to qualify as human authorship, thereby reducing guesswork in litigation.

Critics caution that statutory amendments may quickly become obsolete due to fast-changing technology, and may struggle to capture the infinite variety of user-AI collaborative workflows (Mendis, 2020).

Administrative Guidelines and Soft Law

Regulatory agencies like the U.S. Copyright Office or WIPO could issue more detailed guidance documents:

Outline best practices for registering AI-involved works. For instance, instruct applicants to disclose the extent of user input, supply evidence of iterative refinements, or clarify whether the user or the AI made key artistic decisions.

Allow partial registrations, where only the user’s recognized contribution is covered.

Emphasize disclaimers where the AI’s autonomous role overshadowed any human authorship, leading to no registrable protection.

While not as binding as legislation, such guidance could gradually harmonize real-world practices and reduce confusion for creators, platform developers, and the public.

Contractual Solutions

Given that much AI art is generated through proprietary platforms, many disputes might be forestalled by private contracts. If a platform’s TOS clearly states:

“Users own all rights in the final images, provided they supply the creative direction, but the platform retains a license for promotional or internal training purposes,”

then, in practical terms, the user can act as if they hold copyright, unless a third party challenges that ownership externally (Thornburg, 2021). This approach, though not guaranteeing universal legal recognition, does reduce friction for typical commercial or personal uses of AI-generated art.

Alternative Legal Protections

A handful of scholars suggest that current copyright law may be ill-suited for machine-created works, advocating for alternative protection regimes:

Sui Generis Intellectual Property: A specially crafted legal category offering limited protections for AI-generated outputs, focusing on the person who curated or financed the generative process (Anderson, 2022).

Neighboring Rights: Some jurisdictions have introduced “neighboring rights” that protect entities like record producers or broadcasting organizations without calling them “authors” per se (Bird & Mendis, 2020). A similar concept could apply to AI outputs.

Attribution and Transparency Requirements: Even if no copyright arises, laws or industry norms could mandate identifying AI’s role in creation, ensuring the public is aware of how the piece was generated (Galanter, 2016).

Future Directions and Implications for Creativity

Impact on Artists and Cultural Industries

Artists often grapple with AI tools that can replicate or blend styles, raising fears about losing control over their distinctive “voice” (Anderson, 2022). If user prompts yield near-perfect imitations of a painter’s technique, the line between homage, fair use, and infringement becomes blurred. The question of user authorship also ties into broader socio-economic debates: do we devalue creative labor if we assert that minimal prompting confers full ownership? Or do we stifle creativity if we deny any protection to these emergent works, effectively flooding the market with free, user-generated but AI-rendered art?

In positive terms, many artists and creators see AI as an empowering tool, offering new aesthetic horizons and the potential to accelerate concept exploration. If legal frameworks provide clarity—affirming that genuine user input meets authorship requirements—then professional artists might confidently adopt AI in their workflow. Conversely, an overly restrictive approach might hamper such innovation, restricting the synergy between human imagination and AI-based assistance (Thornburg, 2021).

Evolving Judicial Perspectives

As AI art becomes commonplace, courts will inevitably confront disputes over alleged infringements, ownership conflicts, and authenticity claims. Judicial decisions will shape how strictly the “human authorship” rule is interpreted. Key questions might include:

What evidence can users provide to prove they meaningfully directed the AI?

Does the unpredictability of AI outputs negate claims of user authorship?

Might disclaimers or logs from the AI platform showing user prompts and iteration cycles serve as proof of creative input?

Over time, case law may articulate bright-line standards or remain fact-intensive, decided on a case-by-case basis. If the latter, unpredictability could linger for creators and developers alike.

Ethical and Philosophical Considerations

Beyond legal formalities, the question of “Who Owns Creativity?” in the AI era touches on philosophical debates about what it means to create art. If an AI’s capacity to surprise or innovate approaches that of a human artist, do we expand our definition of authorship to include human-AI teams? Or do we insist that the intangible qualities of human consciousness remain the sine qua non of genuine creativity?

Some ethicists argue that giving too much legal recognition to AI-driven outputs might undermine the significance of human creativity, effectively turning creative endeavors into a series of automated manipulations. Others believe that humans remain integral precisely because they conceptualize the prompts and select the final works, thus injecting personal vision into the machine’s generative apparatus (Manovich, 2019). These broader reflections will inevitably inform how societies and lawmakers respond to the new wave of creative AI.

Conclusion

This paper has explored the specific question of whether user inputs—particularly textual prompts and iterative refinements—can ground a valid copyright claim in AI-generated art. Traditional copyright principles hinge on human originality, generally excluding purely machine-generated outputs from protection. Yet as advanced AI systems proliferate, the nature of user involvement runs the gamut from cursory instructions to deep, iterative co-creation. At some point on this continuum, the user’s direction and editorial control arguably satisfy the minimal creativity threshold required for authorship. Precisely locating that threshold is both legally and practically complex.

Multiple legal theories offer frameworks: the tool analogy, joint authorship, a complete rejection of AI authorship, or partial protection limited to demonstrable human contributions. Real-world outcomes currently hinge on disclaimers in terms of service, early administrative refusals, and the sparse set of relevant case law. Courts and policymakers have only begun grappling with these issues, leaving a patchwork of uncertainties. Moving forward, we may see increased reliance on revised statutes, administrative guidelines, or specialized legal regimes to clarify how AI-generated works are treated. Moreover, the evolution of user practices—more sophisticated prompt engineering, multi-stage curation, and combination of AI outputs with human-driven editing—will likely shape the debate about how “authorship” is established in the eyes of the law.

Ultimately, the question of “Who Owns Creativity?” in the AI era cuts to the heart of how societies value and protect novel expressions. On one hand, conferring robust rights to the user behind the prompt can fuel innovation and reward those who skillfully guide AI systems. On the other, artificially broadening authorship may undermine the principle that copyright exists to protect uniquely human invention. Balancing these competing concerns will require an ongoing dialogue among legal scholars, technologists, artists, and policymakers, ensuring that the law evolves in tandem with the extraordinary capabilities of emerging AI technologies.

Acknowledgments

I would like to thank my research mentor, Dr. Zuri Elysium, for her unwavering guidance, extensive feedback, and inspiring enthusiasm throughout the development of this paper.

References

Anderson, L. (2022). Copyright challenges in the era of creative machines. Journal of Intellectual Property Studies, 14(2), 191–209.

Barrat, R. (2018). The AI art at Christie’s and what it means for the future of creativity. Artnet News. Retrieved from https://www.artnet.com

Bird, R., & Mendis, D. (2020). Policy responses to AI-generated works: A comparative analysis. International Review of Intellectual Property and Competition Law, 51(3), 15–38.

Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53 (1884).

Carroll, M., & Zheng, N. (2021). Fixation in the digital realm: Reconceptualizing ephemeral expression. Fordham Intellectual Property, Media & Entertainment Law Journal, 31(2), 201–228.

CDPA (Copyright, Designs and Patents Act). (1988). Section 9(3). United Kingdom.

Collins, A. (2019). Artificial intelligence and the tool paradigm: Aligning old doctrines with new technologies. Stanford Technology Law Review, 22(1), 131–162.

Elgammal, A., Liu, B., Elhoseiny, M., & Mazzone, M. (2017). CAN: Creative adversarial networks generating “art” by learning about styles and deviating from style norms. Proceedings of the 8th International Conference on Computational Creativity, 96–103.

European Commission. (2020). White paper on artificial intelligence. Brussels: European Commission.

Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991).

Galanter, P. (2016). Generative art theory: A systems approach to aesthetics. Leonardo, 49(2), 183–189.

Gervais, D. (2020). The machine as author. Iowa Law Review, 105(5), 2053–2106.

Goldstein, P. (2017). Goldstein on copyright (4th ed.). Wolters Kluwer.

Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., … Bengio, Y. (2014). Generative adversarial nets. Advances in Neural Information Processing Systems, 27, 2672–2680.

Kingma, D. P., & Welling, M. (2014). Auto-encoding variational Bayes. International Conference on Learning Representations. Retrieved from https://arxiv.org/abs/1312.6114

Karras, T., Laine, S., & Aila, T. (2019). A style-based architecture for GANs. IEEE Conference on Computer Vision and Pattern Recognition, 4401–4410.

Leval, P. (2022). Minimal creativity and the law: Refining thresholds in AI era. Harvard Journal of Law & Technology, 35(1), 221–239.

Manovich, L. (2019). AI aesthetics. AI & Society, 34(4), 753–768.

McCormack, J. (2007). Open problems in evolutionary music and art. In J. Romero & P. Machado (Eds.), The art of artificial evolution (pp. 205–214). Springer.

Mendis, D. (2020). The copyright conundrum of computer-generated works: A critical perspective. European Intellectual Property Review, 42(2), 87–98.

Mordvintsev, A., Olah, C., & Tyka, M. (2015). Inceptionism: Going deeper into neural networks. Google AI Blog. Retrieved from https://ai.googleblog.com/

Naruto v. Slater, 2018 U.S. App. LEXIS 34116 (9th Cir., Apr. 23, 2018).

OpenAI. (2022). DALL·E terms of service. Retrieved from https://openai.com

Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., & Sutskever, I. (2021). Learning transferable visual models from natural language supervision. Proceedings of the 38th International Conference on Machine Learning. Retrieved from https://arxiv.org/abs/2103.00020

Ramesh, A., Pavlov, M., Goh, G., & Gray, S. (2021). Zero-shot text-to-image generation. International Conference on Machine Learning. Retrieved from https://arxiv.org/abs/2102.12092

Rombach, R., Blattmann, A., Lorenz, D., & Esser, P. (2022). High-resolution image synthesis with latent diffusion models. IEEE Conference on Computer Vision and Pattern Recognition. Retrieved from https://arxiv.org/abs/2112.10752

Rosen, J. (2022). Prompts as authorship: Charting a path for AI-generated art. Berkeley Technology Law Journal, 37(1), 102–145.

Samuelson, P. (1986). Allocating ownership rights in computer-generated works. University of Pittsburgh Law Review, 47(1), 1185–1226.

Thornburg, M. (2021). Midjourney, DALL·E, and the uncertain landscape of AI art ownership. New Media & Society, 23(7), 1561–1578.

U.S. Copyright Office. (2019). Compendium of U.S. copyright office practices (3rd ed.). Washington, D.C.