The Copyright Algorithm: Does AI-generated Ghibli break the law or open new possibilities?

OpenAI’s ChatGPT has opened up a new dreamy world of AI-generated Ghibli art – inspired or copied from legendary Japanese animator Hayao Miyazaki’s Studio Ghibli style. It didn’t take it long for Ghibli art to become viral, with politicians to celebrities to netizens experimenting with their Ghibli-fied avatars and sharing for all to see. Personal photos, memes, movie scenes, famous art – all have made it to the AI-generated Ghibli artscape.

However, even as this trend spreads like wildfire, there are growing voices of concern regarding the serious copyright and privacy muddle that AI-generated Ghibli art can leave in its wake. Miyazaki himself has thundered against AI’s intrusion into art, famously branding the idea of AI-generated art as “an insult to life itself”. This powerful sentiment underscores the profound unease gripping many artists as algorithms aggressively encroach on traditionally human creative domains.

In a crucial move to confront the escalating legal uncertainties suffocating the intersection of Artificial Intelligence (AI) and copyright, the Japan Copyright Office (JCO) has released a pivotal document. The document, starkly titled “General Understanding on AI and Copyright in Japan – Overview”, provides an initial outline of the office’s perspective on this incendiary issue. The Legal Subcommittee under the Copyright Subdivision of the Cultural Council officially declared the “General Understanding”, with the JCO, operating under the Copyright Division of the Agency for Cultural Affairs, Government of Japan, wielding the pen as publisher. The public was granted access to this potentially game-changing overview in May 2024.

The “General Understanding’s” core mission was to inject clarity into the murky waters of how the existing Japanese Copyright Act should be interpreted and, more importantly, applied within the context of rapidly advancing and often unpredictable AI technologies. However, the JCO, with a note of caution, emphasized that the guidelines within the document are not legally binding and should not be considered definitive legal advice for the chaotic realm of specific generative AI technologies. This publication arrives not a moment too soon, as stakeholders across the entire spectrum – AI developers hungry for data, copyright holders fighting for control, and bewildered users caught in the crossfire – grapple with the unprecedented and often terrifying challenges AI hurls at traditional copyright frameworks.

The Million-Dollar Question: Learning or Looting?

When an AI model is unleashed on a dataset crammed with copyrighted artistic styles, like the distinctive magic of Studio Ghibli, how does one even begin to define the razor-thin line between legitimate ‘learning’ and outright ‘replicating’ in the eyes of copyright infringement?

The distinction between learning and replicating hinges precariously upon originality and transformation. Meher Patel, Founder of Hector, explains, “Learning occurs when an AI model analyzes various artistic elements, such as color palettes, composition, and character design, to produce outputs that integrate these influences into a uniquely new creation. Replication, however, implies producing outputs that closely mimic specific protected artistic expressions, lacking sufficient transformative value. To assess infringement, it’s crucial to evaluate whether the AI-generated content presents a new, original interpretation or if it substantially mirrors the protected creative elements of the original artwork.”

Abhinav Jain, Co-Founder and CEO of Almonds AI, injects a dose of stark reality, “Learning means that the AI is analyzing patterns and styles at a fundamental level like understanding colors, compositions, and structures but not directly copying any specific image. Replicating, however, becomes a grave concern when the AI produces works that are too similar to the original copyrighted material. The fine line often comes down to whether the AI’s output is so close to the original work that it risks confusing the source material with something new. There’s a grey area here, a legal and ethical quagmire, and it’s a conversation we desperately need to have more openly within the tech and creative communities.”

The Ethical Tightrope: Who Guards the Guardians?

In the unsettling absence of clear legal guidelines, what profound ethical responsibilities do AI developers shoulder regarding the use of copyrighted artistic styles in their insatiable training data?

“AI developers carry an ethical responsibility to fiercely respect and genuinely acknowledge artists’ intellectual contributions,” asserts Meher Patel. “Even without explicit legal frameworks, developers should proactively seek permissions, transparently disclose their often-secret training datasets, and credit original artists or studios when their styles significantly influence AI-generated outputs. Implementing robust internal guidelines that promote transparency, informed consent, and fundamental fairness can help developers ethically navigate copyright complexities and, crucially, cultivate trust among increasingly skeptical creative communities.”

Abhinav Jain stresses, “While there’s still a distressing lack of legal clarity, AI developers have a non-negotiable responsibility to act with profound respect for the original creators. Developers should be transparent about the immense amounts of data they use to train these voracious models and ensure that they’re not violating the very spirit of creative work. If AI is essentially learning on the backs of copyrighted material, the original creators should unequivocally benefit, either through direct collaboration or by scrupulously respecting their intellectual property rights. The ethics here boil down to simple fairness ensuring that creators aren’t exploited and crushed by the very technologies meant to advance creative potential.”

The Unforeseen Apocalypse: Will AI Art Destroy Art Itself?

What are the potential unintended and potentially catastrophic consequences of AI-generated art flooding the already fragile art market and irrevocably altering the broader cultural landscape?

“AI-generated art carries multiple potential of unintended consequences, each one more alarming than the last,” observes Meher Patel. “It could lead to a devastating market saturation, drastically complicating the valuation of human-created artworks by sheer volume, increasing availability, and fatally reducing perceived uniqueness. Moreover, it might inadvertently and tragically marginalize original artists, whose distinctive styles, honed over years, may be diluted or brutally undervalued in this new digital free-for-all. On a cultural level, a widespread reliance on AI-generated outputs could homogenize creativity itself, insidiously undermining diversity in artistic expression. Conversely, in a flicker of hope, it might also inspire entirely new artistic movements, prompting human artists to innovate even further to desperately differentiate themselves, thus potentially reshaping cultural narratives.”

“AI is undeniably revolutionizing creativity, but it also drags with it some deeply troubling unintended consequences,” warns Abhinav Jain. “In the art world, there’s the very real potential for a swift devaluation of traditional art forms if AI-generated pieces, churned out by algorithms, flood and drown the market. For example, AI can produce virtually limitless amounts of art at lightning speed, a terrifying prospect for artists who pour months, even years, into crafting a single, unique piece. It could disastrously shift the demand towards soulless speed over genuine artistry. Culturally, it also forces us to confront the chilling question: if machines can now generate art, does it fundamentally dilute the precious human connection to creativity? Ultimately, AI-generated art will violently challenge traditional ideas of originality, ownership, and the very concept of value in art, but, perhaps optimistically, it will also unlock entirely new and unforeseen possibilities for expression. That being said, AI can be an incredibly powerful tool for artists, a digital assistant rather than a cold replacement. The key, as always, is finding a delicate balance—where AI assists and augments creativity, rather than overshadowing and ultimately obliterating it.”

The Line in the Sand: Style or Theft?

Given the inherent nature of AI models to learn and reproduce patterns, not churn out exact copies, how can we possibly define any defensible threshold between ‘style replication’, a murky copyright issue, and ‘direct asset theft’, which is outright piracy?

Meher Patel posits, “The threshold between style replication and piracy hinges entirely on the specificity and substantiality of the reproduced content. Direct asset theft occurs when exact, identifiable elements such as specific characters, compositions, or narratives are duplicated without any attempt at transformation. Style replication, conversely, involves merely emulating general artistic techniques or aesthetics without directly copying distinct, protectable expressions. Defining this incredibly slippery threshold requires painstakingly evaluating the degree of genuine creativity, demonstrable originality, and transformative input present in any AI-generated work, ensuring an incredibly delicate balance between rigorously protecting original creators and simultaneously fostering both technological and artistic innovation.”

According to Abhinav Jain, “This is where things get incredibly tricky, a legal and ethical minefield. If an AI generates something broadly in the style of Ghibli, is that fundamentally different from a human artist doing essentially the same thing? Technically, perhaps. But practically, it all hinges on the final output. A solid test is: Does the AI-generated work genuinely feel like an independent creation, standing on its own merits, or does it chillingly look like it was simply ripped straight from an artist’s existing portfolio? If an AI generates a piece that slavishly resembles an actual frame from a Studio Ghibli film, that’s almost certainly asset theft. If, however, it produces something tangentially inspired by Ghibli’s characteristic soft colour palettes and whimsical details that leans closer to style replication.”

Jain believes that the safest and most responsible approach would be to meticulously design AI models from the ground up to actively avoid overfitting to specific artists’ works – which inherently means setting firm limits on just how closely the AI is even permitted to mimic a single reference. “We urgently need far better and more robust guardrails to ensure AI remains a powerful tool for inspiration and augmentation, rather than an automated and ultimately destructive shortcut for blatant imitation,” he concludes.

Also Read: Scraping, Suing and Settlements: The Copyright battle between AI and Media