Recently, ChatGPT’s latest ability to generate Ghibli-style anime images has ignited an internet frenzy, inspiring countless memes and creative reinterpretations of familiar visuals, to leaving copyright questions unanswered. As artists, news publishers, and musicians launch lawsuits alleging copyright infringement, it is vital to examine the legal framework that governs this digital creativity—especially within the EU.
In this article, I will explore how EU copyright laws intersect with the phenomenon of “ghiblifying” images, look into the controversies surrounding OpenAI’s latest model and share some of my own experiences and examples.
The Ghibli Phenomenon
Studio Ghibli has long been celebrated for its hand-drawn masterpieces, such as My Neighbor Totoro and Princess Mononoke. These films resonate deeply due to their aesthetics – detail, emotive character designs, and a harmonious blend of color and light. The AI community’s fascination with replicating this iconic style has led to the phenomenon of “ghiblifying” images, where the distinctive aesthetics of Studio Ghibli are reinterpreted through the lens of modern technology. Despite its creative appeal, ChatGPT’s Ghibli-style feature has sparked both praise and controversy. Initial reports falsely suggested it would be limited to paid users, but it remains free for everyone—upholding the democratization of artistic tools coherent with OpenAI’s policies.
Why is the model so good?
Generative AI programs have long relied on producing outputs in response to user prompts. Previous generations of image generators predominantly employed diffusion models. These models operate by gradually transforming random, noisy data into coherent images through iterative refinement. Although effective, this approach often required users to input lengthy prompts to achieve desired results.
The latest update to ChatGPT. Instead of using diffusion models, now uses an autoregressive algorithm. This method treats images more like language. Just as the model predicts the most likely words in a sentence, it breaks down images into “tokens” and predicts individual visual elements step by step.
This tokenisation allows the algorithm to separate and control distinct features of an image, aligning them closely with the descriptive words in a user prompt. Consequently, ChatGPT can generate images with higher precision, replace or modify specific elements while retaining the overall structure, and even overcome longstanding issues such as generating legible text within images. Moreover, integrating image generation within a large language model means the AI can draw on an immense reservoir of contextual knowledge. Users no longer need to detail every aspect; a simple reference to Studio Ghibli is enough for the model to understand and recreate its style.
Hayao Miyazaki’s Critique
No discussion of Ghibli-style AI images would be complete without addressing the concerns voiced by Hayao Miyazaki, the legendary director behind many of Studio Ghibli’s masterpieces. A resurfaced video from 2016 shows Miyazaki passionately denouncing AI during a studio demonstration, stating, “I would never wish to incorporate this technology into my work at all. I strongly feel that this is an insult to life itself.” His words, although delivered years ago, have taken on renewed significance in light of current debates.
Controversial Imagery and Misuse
Among the most contentious examples of AI-generated content was an image shared by the White House on X (formerly Twitter). The post featured an AI-generated depiction of a felon—defined as a person convicted of a serious crime—being handcuffed by a US immigration officer. This image, aside from its political overtones, sparked immediate condemnation for its overtly racist implications.
From a governmental perspective, such imagery is not only a misuse of AI but also a stark violation of the ethical standards expected of public institutions. Using AI to propagate racially charged stereotypes is irresponsible and, frankly, unacceptable. The dissemination of such content undermines public trust and fuels divisive narratives, which should have no place in official communications.
OpenAI’s Controversy
Back to OpenAI’s foray into Ghibli-style image generation and the latest chapter in an ongoing saga of copyright disputes. This is not the first time the company has faced legal challenges over its use of copyrighted content. In the past, similar controversies have erupted as artists, news publishers, and musicians have accused OpenAI of unauthorized appropriation of their intellectual property.
EU Legislation in Focus
In the European Union, the legal framework surrounding copyright is notably stringent. The Information Society Directive (2001/29/EC), for example, lays the foundation for protecting the rights of digital creators, which we have covered in our article on AI, Copyright, and the Future of Content Creation: How Global Regulations and the EU AI Act Are Reshaping Creators’ Rights.
More recently, the Digital Single Market Copyright Directive has introduced further measures aimed at tightening the rules around online content sharing. These regulations are designed to protect the creative and economic interests of artists and other content creators by ensuring that digital platforms take greater responsibility for preventing unauthorized use of copyrighted material. In the context of AI, this means that if an algorithm reproduces a “substantial part” of a copyrighted work – and it could be liable under EU law.
The complexity deepens when considering that the EU’s approach diverges significantly from the United States’ concept of fair use. Whereas U.S. law allows for broader transformative uses under the fair use doctrine—covering areas like satire, commentary, and meme culture—the EU takes a more conservative stance, offering less flexibility. This difference is at the heart of the current legal debates, as stakeholders on both sides argue whether AI-generated art constitutes an innovative reinterpretation or a direct infringement of existing copyrights.
OpenAI’s Response and the Road Ahead
Faced with a barrage of lawsuits and mounting public scrutiny, OpenAI when questioned by AFP about potential infringements on Studio Ghibli’s intellectual property, a company spokesperson emphasized that the model is still in a fine-tuning phase. OpenAI maintains that while it prevents the replication of styles tied to individual living artists, it does allow for broader studio styles—a distinction intended to foster creative fan reinterpretations without crossing legal boundaries.
Furthermore, OpenAI is actively lobbying both the White House and the US Congress to address the complex issue of copyrighted content used by AI companies. Their goal is to expand the scope of fair use to include such innovative applications—a move that would align digital content regulations more closely with modern technological capabilities. However, the success of such lobbying efforts remains uncertain, particularly given the conservative approach to copyright that prevails in the EU.
Experimentation and Personal Reflections
I tested the system with my own images and was truly impressed with the results – it captured the essence of the Ghibli style remarkably well, allowing me to fine-tune details without sacrificing the overall image integrity. I tested working with image generating based on a written poem, and the text seems correct and fluent. At this point, this is the best model for replicating text. The image quality is good and prompting is exceptionally easy. However, these encouraging outcomes are shadowed by concerns over long-term copyright issues, as the continued use of such technology may eventually spark significant legal challenges regarding originality and intellectual property rights.
In this ongoing dialogue, the key is not to stifle innovation but to channel it responsibly. As we move forward, let us ensure that the transformative power of AI enhances our creative landscape, enriches our cultural heritage, and upholds the fundamental rights of every creator in the digital age.
