Animating The Line - Studio Ghibli, AI-Generated Art, And The Crosscurrents Of Copyright And AI Law
Author: LawSphereX
Category: Technology, Cyber & Privacy
Published:
Updated:
ANIMATING THE LINE: STUDIO GHIBLI, AI-GENERATED
ART, AND THE CROSSCURRENTS OF COPYRIGHT AND AI
LAW
Anoushka Mathur & Vagisha Gupta, BCom LLB (Hons.), O.P. Jindal Global University
ABSTRACT
The rapid growth of generative AI, like the recent ChatGPT4o, has ignited widespread debates over copyright and moral rights in today’s digital age. This paper aims to examine the legal and ethical complexities arising from AI generated art, especially when it mimics the distinctive styles of artists, like Studio Ghibli. Copyright laws, such as India’s Copyright Act, 1957, protect tangible expressions of ideas. However, they fail to address the replication of styles of artists, leaving notable creators threatened by unauthorised derivative use. This analysis explores whether AI outputs are infringing derivative works or merely transformative fair use, referencing landmark cases like Andy Warhol Foundation v Goldsmith and includes the challenges faced by creators while enforcing copyright against amassing AI generated content.
This paper also discusses moral rights concerns, using Studio Ghibli’s pacifist ethos being utilised to generate contradictory militarised images and thus distorting creator’s original intent. Jurisdiction enforcement hurdles and platform liability under EU AI Act and India’s Data Protection Act are also scrutinized to reveal gaps in implementation of accountability. To round up, the paper includes recommendations that include international cooperation, opt-in mechanisms for copyrighted data and legal reforms that adapt to changing scenarios to balance innovation and protections. Ultimately, the paper highlights the urgent need for new and evolved IP frameworks to address AI’s disruptive impact, especially and most importantly on artistic ownership and integrity.
Introduction: If your social media feed in the past couple of weeks has not been completely bombarded with sweet-looking animated faces of people that are vaguely familiar, it can be alleged you are living under a rock. With the introduction of a new model of ChatGPT called ChatGPT4o, users have gained access to the realm of AI image generation that few AI platforms offered before. This new model of OpenAI seemed trained in the Studio Ghibli art style, allowing users to turn their pre-existing family photos into cartoon versions of themselves, making for new profile pictures across the world.
However, this seemingly innocent trend online revealed multifaceted controversies and glaring loopholes in our legal systems- from privacy rights to copyright issues, highlighting that the contemporary world isn’t prepared to handle the onslaught of media that Generative AI is threatening to unleash. With these unrestricted image generation powers that certain AI platforms have granted users, come issues that have been leading discussion on generative AI since its inception- probable copyright infringement of information used by developers to train AI, moral rights of the authors of such media, including autonomy over where their creations are used and finally, whether in cases of generated art, can an art style be truly copyrighted by its producers. Are styles copyrightable?
The question of whether an artistic style can be copyrighted is the core issue surrounding tensions between artistic originality and creative replication and derivation. When watching a Studio Ghibli production, you see bright colours, water-colour inspired backgrounds and fantastical architecture. This is the broader artistic style of their work which expresses itself in particular scenes, images and illustrations. However, under current copyright jurisprudence, including the Indian Copyright Act 1957, only specific and tangible expression of an idea is protected. This implies that the ‘aesthetic’ of Ghibli artwork, including repetitive colour patterns, mood settings, etc, is unprotected and remains in the public domain; what is protected from reproduction are specific renditions of the characters or a particular frame from the movies.
The latter is fixed and human-made art that cannot be reproduced directly, without the author’s permission. However, with AI tools that are suddenly able to mimic these styles without directly reproducing the original images, the distinction between the style and its expression is becoming increasingly blurred and the artist rights over their art are becoming more and more narrow. It raises the question that when AI is used to generate content that
mimics distinctive art styles of artists, does this amount to illegal copying or merely inspiration for derivative works? How AI generates these images: For this, we must look at the basics of what AI does. To train AI to give reasoned and accurate responses, developers must download and store data (images, text, etc. ). This counts as making copies, which is something that, by law, only the owner of the copyright may do. In strict copyright jurisdictions, like the US, even if the data is not directly linked to monetary benefits, downloading is sufficient to attract liability, with fair use defences falling short to save the developers. 1 Fortunately, AI only uses these data sets to learn patterns, and not pure reproduction. It uses these patterns to generate responses.
For example, ChatGPT4o learnt that Studio Ghibli uses soft colours and round shapes in their art and uses this knowledge to create or recreate uploaded images. Question of whether this is fair use of copyrighted material remains unanswered, with many jurisdictions having pending litigation regarding the same. Many courts, like in the US, use the principle of non-expressive use to justify fair use of copyrighted material. 2 This implies, that AI, in its training and learning process, does not learn the creative parts of a work, but simply the constituents of the data that can commonly be used to identify a particular concept. For example, it will identify that a bedroom consists of a bed, table and chair and use that to distinguish between a bedroom and a beach.
Such non-expressive elements of works, like facts or ideas, aren’t copyrightable, but for AI, they are useful to learn about data stored for it. Are the AI images derivative works or wholly new transformative outputs? Derivative works, unauthorised reproductions of copyrighted art, are identified as being remarkably like the initial protected work. This is something that the courts have found difficult to establish as AI rarely replicates a single image that it was trained upon. 3 The resemblance that AI outputs do show, can be argued to classify as transformative fair-use, especially if it adds new expression or meaning. 4 Defenders of AI art argue that the outputs are novel work
1 Jenny Quang, Does Training AI Violate Copyright Law?, 36 BERKELEY TECH. L.J. 1407 (2021).
RAND
Corporation, Artificial Intelligence Impacts on Copyright Law,
PEA3243-1
(2024), https://www.rand.org/pubs/perspectives/PEA3243-1.html 3 Eleonora Rosati, Generative AI and Copyright: Principles, Priorities and Practicalities, 18 J. Intell. Prop. L. & Prac. 841 (2023). https://academic.oup.com/jiplp/article/18/12/841/7331468?
4 Ibid.
with no substantial similarity to any one work, but broad patterns turned into something newan inspiration if you will. 5 Though the court has yet to decide on whether AI outputs count as either, the US Supreme Court in Andy Warhol Foundation v. Goldsmith,6 with the weight of 6 judges warned that allowing any aesthetic transformation to count as fair use would undermine the right to create derivative works of the copyright owner. This analysis is also crucial in determining whether AI generating “style” application infringes on the rights of original creators. If that is not the case, it risks not mere infringement on the economic rights of the creator, but also its moral rights- which entitle the artist to determine where their art will be used, etc.
Studio Ghibli and their Moral Outlook: The co-creator of Studio Ghibli, Hayao Miyazaki, has famously expressed his outlook on AI art, calling it an “insult to life itself”. His belief stems from the inability of AI of replicating the genuine human experience and emotions, which lead to the creation of authentic art- full of soul and humanity. Throughout his creations, he promotes the anti-war and violence agenda, showing soldiers as misled fighters at best, and sexual predators at worst. 7 However, this social media trend has led to infamously militarised governments and departments Ghibli-ing their photos through AI, including the Israeli Defense Forces, which outrightly goes against Studio Ghibli’s ethos.
Other photographs by social media users included the AI anime image of the destruction of the historic Babri Masjid in 1992 by Hindutva rioters; another violent rendition that went against the foundation principles of Studio Ghibli’s arts. Many laws, both national and international, grant the moral right to authors to prevent any distortion, mutilation or other modification that can harm their honour or reputation. This covers those alterations that can change the work’s artistic message or meaning. 8 These moral rights have gained legal grounding in India through section 57 of the Indian Copyright Act. This provision is closely linked with Article 6bis of the Berne Convention, to which there are
181 signatories. These renditions of scenes in the classic Studio Ghibli style can be argued to
5 Ibid.
6 Andy Warhol Found. for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508 (2023).
7 Sahana Venugopal, Is OpenAI’s Studio Ghibli-style AI image generator undermining copyright law?, THE HINDU (Apr. 11, 2025) https://www.thehindu.com/sci-tech/technology/is-openais-studio-ghibli-style-ai-imagegenerator-undermining-copyright-law/article69438317.ece
8 E-Justice India, Author's Moral Rights: Copyright Law in India, E-JUSTICE INDIA (Mar. 1, 2021),
https://www.ejusticeindia.com/copyright-law-in-india-authors-moral-rights/
violate the right to integrity of the copyright holder, with the pacifist stance of the creators being close to mocked. Ultimately, who owns the output? The question remains- who owns this AI output? The developer or the creator? Is this output copyrightable at all? In India, copyright protection is granted to original works created by a human author. More specifically, for computer generated works, the “author” is the person who causes the work to be generated. 9 The question remains whether AI outputs can be equated to computer generated works, especially as AI can generate content without much human intervention. In fact, in 2020, AI RAGHAV, a painting app, was granted co-authorship with the owner of the app for an image of a sunset that had striking similarities to Van Gogh’s ‘Starry Night’ and its painting style and colours.
However, the Copyright office later recalled the copyright granted to the AI app, reinforcing that only natural persons were allowed to hold copyrights within India. It is also seen that the Indian judiciary have maintained that authorship requires human creativity. 10 In 2019, the Delhi High Court denied copyright protection to a computer-generated list, as human intervention was absent. 11 It is arguable however, if a user provides creative prompts with significant editing to the output, they may still be granted authorship of that work. With this stance, mere prompt based generative AI output stands without clear protection under copyright law and ownership rests on involvement and editing by the author, a purely fact-based determination, leaving wide room for litigation.
This ambiguity highlights the gaping hole between technological advancements and the grasping power of legislature around the world, even though basic groundwork has been laid in some jurisdictions like the EU. India, too, has seen the Parliamentary Standing Committee’s 161st report on Review of the Indian Intellectual Property Rights Regime in 2021, which recommended the creation of a separate category of rights for AI and related innovations. This effort has yet to bear fruit.
9 The Copyright Act, No. 14 of 1957, § 2(d)(vi)
10 Srishti Ojha, ChatGPT, AI-Generated Content and Copyright: Who Owns the Output?, INDIA TODAY (Sept. 22, 2023), https://www.indiatoday.in/law/story/chatgpt-ai-generated-content-copyright-ownership-complexitiesindia-2439165-2023-09-22.
11 Satish Kumar v. Khushboo Singh & Ors, 89/2019, (Del. High Ct. Oct. 14, 2019)
Practical Difficulties in Enforcing Copyright Against AI-Generated Lookalikes The burgeoning of AI-generated content presents enforcement challenges, as unlike traditional copyright infringement where identifiable human actors reproduce protected works, here AIgenerated imitations are created autonomously by algorithms trained on a vast amount of data which is often taken without authorization. This makes AI capable of generating unique, personalized content with minimal human intervention. This represents a significant departure from earlier models of internet content which merely indexed existing information but now, with AI tools synthesizing new outputs directly, it poses significant challenges for copyright holders like Studio Ghibli. The sheer volume of these AI-generated derivatives, like in the case of Studio Ghibli's imitations, complicates enforcement.
This kind of digital content is highly scalable12 as duplicating and sharing it takes little to no effort. Copyright holders then have an impossible task of monitoring and issuing takedowns of thousands of AI-generated images, many of which may not be direct copies but stylistic imitations that still exploit Ghibli's creative essence. Furthermore, because of the global and interconnected nature of the internet, jurisdictional issues arise as even if Ghibli successfully enforces its rights in Japan, infringing content may persist in other countries where Ghibli might have weaker IP protections. Platform Liability The question remains whether AI platforms that enable this replication (such as ChatGPT) should bear liability for copyright infringement.
The most prominent law regulating AI is the EU AI Act, under which general purpose AI platforms like ChatGPT are subject to transparency and copyright compliance obligations. Article 53 of the Act mandates that the providers of such models implement a policy to respect EU copyright law, including honouring opt-outs from text and data mining exceptions and publish a sufficiently detailed summary of training data used. However, the act itself doesn't establish any liability structure. Instead, liability for copyright infringement may be pursued under the proposed AI Liability Directive- which includes a "rebuttable presumption of causality" in cases of harm caused by AI systems. This shifts the burden of proof in certain circumstances to AI developers, especially if they failed to 12 R. A.
Aswin Krishna, Unraveling Intellectual Property: A Study on the Transformative Role of AI and Digital Innovations, 7 Int’l J. L. Mgmt. & Hum. 2661, 2666 (2024).
meet a duty of care, which could include things such as ignoring copyright restrictions in training data. 13 So, if an AI model is demonstrably trained on Ghibli’s copyrighted works without permission, and its outputs closely resemble protected films, the studio could argue that the platform failed to implement safeguards against infringement. However, under another EU law i. e. , the Digital Services Act (DSA), platforms may claim "safe harbour" protections if they act neutrally which means that they do not actively encourage infringement and have safeguards to prevent it. 14 If ChatGPT merely provides a tool without explicitly promoting Ghibli-style outputs, it may evade liability. In the US, OpenAI has claimed protection under the fair use doctrine, arguing their models are “transformative” and that the training process does not directly expose copyrighted content to users.
Courts assess fair use based on four factors: the purpose of use, the nature of the work, the amount used, and the market effect. For example, in Campbell v. Acuff-Rose Music,15 the court emphasized that transformation—adding new expression or meaning—is central to fair use. However, this argument weakens if the AI-generated output serves the same market function as the original work, as seen in recent pushback from media outlets like the New York Times. From an Indian perspective, Section 52 of the Copyright Act, 1957 includes limited fair use exceptions, such as for research or private use, but lacks clear guidelines on machine-generated content. As the 161st Parliamentary Standing Committee observed, the Act is ill-equipped to handle AI-generated works and recommended exploring patent-style protections for AI inventions.
16 Furthermore, the Digital Personal Data Protection Act, 2023 grants users the “right to be forgotten,” allowing individuals to request the deletion of their data. Yet, once an AI model is trained, it cannot be retroactively “untrained,” raising doubts about the practical enforceability of these rights. The Indian Judiciary has addressed the issue of AI infringing IP rights in the Delhi HC case Anil Kapoor v. Simply Life India, the court issued an injunction against the commercial use of 13 Ana Rački Marinković, Liability for AI-related IP Infringements in the European Union, 19 J. Intell. Prop. L. & Prac. 784, 790 (2024), https://doi. org/10. 1093/jiplp/jpae061.
14 Id. at 791.
15 Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569 (1994).
16 Dep’t-Related Parliamentary Standing Comm. on Commerce, 161st Report, Review of the Intellectual Property Rights Regime in India (July 23, 2021).
AI-generated fake or altered content. 17 Moreover, the US District Court in Matera v. Google Inc. , held that remedial measures such as to prevent copyright infringement and privacy protection must be construed broadly to include "new technologies". 18 This underlying idea for up-keep with the pace of technological advances has been accepted by the Indian Courts as well. However, as the 161st Parliamentary Report suggested, “there is a need to review the provisions of [the Copyright Act] on a priority basis”. 19 Recommendations & The Way Forward OpenAI and Google recently put forward policy suggestions in the US, arguing that limiting AI training on copyrighted content, particularly those influenced by “strict European models”, will damage America's technological advantage and hinder innovation, creating a national security risk by letting rival nations pull ahead in AI development.
20 This debate highlights how global AI regulation will largely depend on two factors: a country’s data protection laws and its broader stance on AI as a strategic priority. AI is seen as a vital sector where if a country is lagging, it could have significant geopolitical consequences in the future, incentivizing nations to adopt for more expansive policies to maintain competitiveness. The argument by Open AI and others have been that building AI requires large amount of data scraping and that includes copyrighted data, but this also has multiple issues. Relying on large amounts of copyrighted material not only makes them liable to legal action in the future which could deter investors but also unchecked data collection risks training AI on flawed or biased material, leading to costly fixes down the line.
To solve this, AI models could be trained upon more finely combed material of perhaps even copyrighted material that can be facilitated by international organisations like the WTO. Moreover, after an opt in/opt out mechanism like the European model, a further enhancement could be made by way of including specific citations of where the material has been taken from, which would be easier if the data was already taken by way of a licensing agreement with the IP holder. Additionally, some restrictions must be
17 Anil Kapoor v. Simply Life India & Ors., CS(COMM) 652/2023, Delhi High Court, September 20, 2023
18 Matera v. Google Inc., No. 15-CV-04062-LHK (N.D. Cal. Aug. 12, 2016)
19 Swati Sharma & Nityesh Dadhich, Guardians of Genius: Securing Tomorrow’s Generative AI via Copyright Protection, Cyril Amarchand Mangaldas Blog (Oct. 23, 2023), https://corporate.cyrilamarchandblogs.com/2023/10/guardians-of-genius-securing-tomorrows-generative-ai-viacopyright-protection/. 20 Virginie Berger, The AI Copyright Battle: Why OpenAI and Google Are Pushing for Fair Use, FORBES (Mar. 15, 2025), https://www.forbes.com/sites/virginieberger/2025/03/15/the-ai-copyright-battle-why-openai-andgoogle-are-pushing-for-fair-use/.
placed on the replication of data as it is to limit laymen from infringing upon the IP holder’s rights. But above all, there is a clear need to demarcate the rights of IP holders and outline the obligations that AI developers must respect. While domestic laws are essential, amendments to international agreements like TRIPs are also required to clarify and streamline obligations on a global level. Conclusion The rise of AI generated content has brought with it a crisis of authorship and ownership, exposing the inadequacy of the existing IP frameworks in addressing the burgeoning reality of Artificial Intelligence. It puts the artists rights at risk, facilitating wide sharing of work resembling the artists’, potentially in places which do not align with the ethos of the creator, as it did with Studio Ghibli.
Companies have been using large amounts of copyrighted and private data stating that corporations should not be held to the same standard as everyone else for illegal conduct. This has been mockingly dubbed as the “Bob Dylan defense” as it mirrors the lyrics of one of his songs – “Steal a little and they throw you in jail / Steal a lot and they make you king. ”21 This double standard does paint a troubling reality as under current laws, tech giants face few consequences for appropriating creative labour, while individual infringers bear the full weight of enforcement. To resolve this, policymakers must construct an adaptive legal framework which holds AI developers accountable without stifling innovation.
This requires oversight against corporate overreach, strict transparency requirements, fair compensation schemes for the creators and international cooperation to fix the gaps in legislation that would not only set clear international standards but also put a halt to any argument from corporations’ side about sociopolitical considerations as a vast majority of the world adheres to such standards. 21 Ashley Belanger, Meta Mocked for Raising “Bob Dylan Defense” of Torrenting in AI Copyright Fight, ARS TECHNICA (Mar. 12, 2025), https://arstechnica. com/tech-policy/2025/03/meta-mocked-for-raising-bob-dylandefense-of-torrenting-in-ai-copyright-fight/.