The Shadow of 'Draw in Ghibli Style': How GPT-4o's Image Generation Trend Reveals AI's Dual Nature

 

With the recent release of OpenAI's GPT-4o image generation model, the prompt "Draw in Ghibli style" has spread like wildfire among users worldwide. Beautiful images resembling the unique aesthetic of director Hayao Miyazaki and Studio Ghibli have flooded social media, inspiring admiration from many. However, beneath the dazzling surface of this new feature lies a deep shadow of copyright controversies and environmental issues. This article explores in depth both the fascinating possibilities of AI image generation and the ethical and environmental dilemmas hidden behind it.

1. The Magic Prompt: Phenomenon and Reaction

The Beginning and Spread of the Trend

Shortly after GPT-4o's image generation capability was unveiled, users quickly discovered that this model could recreate Studio Ghibli's style with astonishing accuracy. Simply instructing it to "Draw in Ghibli style" produced dreamy, detailed images reminiscent of Ghibli works like 'Spirited Away,' 'Howl's Moving Castle,' and 'Ponyo on the Cliff.'

The phenomenon rapidly established itself as a viral trend on social media. Users created unique characters and landscapes that would fit perfectly into the Ghibli universe through their own distinctive prompts, offering fresh enjoyment to many Ghibli fans. Creative prompts such as "Modern city in Ghibli style," "Space exploration in Ghibli style," and "Historical events drawn as Ghibli characters" became particularly popular.

GPT-4o's Technical Leap

While previous generations of AI models could generate Ghibli-style images, GPT-4o's results feel more sophisticated and authentic. Its ability to accurately reproduce Miyazaki's distinctive detailed background descriptions, unique character designs, and soft color schemes has impressed even many experts.

This advancement is due to GPT-4o being a multimodal model that understands text and images more deeply and has learned from more extensive datasets. Particularly in image generation, it can more accurately grasp the user's intent and reproduce not just simple style mimicry but also the core aesthetic principles of that style.

2. The Gray Area of Copyright: Whose Art Is It?

The Core of the Legal Controversy

GPT-4o's Ghibli-style image generation capability raises complex legal questions about the boundaries between artistic style and copyright protection. The key issues are twofold:

First, is artistic style itself protected by copyright? In the United States and most countries, copyright generally protects 'expression' rather than 'ideas,' and abstract style is considered closer to an idea. However, in Studio Ghibli's case, one could argue that Hayao Miyazaki's unique aesthetic is so closely tied to the studio's commercial identity that it goes beyond mere 'style.'

Second, does using copyrighted works to train AI models qualify as 'fair use'? OpenAI has stated that its models have learned from billions of images and texts but has not disclosed specific dataset sources. If GPT-4o learned from Ghibli works without explicit permission, this could be grounds for legal dispute.

Studio Ghibli and Creators' Perspectives

As Ghibli-style image generation has become popular, Studio Ghibli has not issued an official statement, but director Hayao Miyazaki has previously expressed negative views on AI technology. In a 2016 interview, he criticized AI-created animation as "an insult to human suffering."

In the broader creative community, concerns have been raised that AI-generated images could threaten the livelihoods of millions of illustrators and animators. Particularly, artists who have been drawing Ghibli-style illustrations express frustration at seeing their work instantly mimicked by AI.

The US Copyright Office Position

The US Copyright Office stated in its 2023 guidelines that "works entirely generated by AI cannot receive copyright protection." However, it included an exception clause that copyright protection is possible if a human has made substantial creative contributions to the AI process.

These ambiguous standards further complicate legal judgments on AI-generated content. There are no clear answers as to whether the prompt input by a user can be recognized as a "substantial creative contribution," and if so, how to address potential copyright issues that AI may have infringed in the image generation process.

3. The Hidden Environmental Cost: Digital Carbon Footprint

Enormous Energy Consumption

The operation of large-scale AI models like GPT-4o consumes tremendous energy that many people fail to recognize. Image generation features in particular require much more computational power than text generation, which translates to higher power consumption.

According to research, training large language models (LLMs) like GPT-4o requires thousands of megawatt-hours (MWh) of electricity, equivalent to the annual energy consumption of thousands of households. Additionally, considerable energy continues to be consumed during the inference stage when the model is actually in service.

As Ghibli-style image generation became popular, OpenAI CEO Sam Altman mentioned that "GPUs are melting," implying severe server load. This suggests that the trend has caused much higher energy consumption than usual.

Carbon Emissions and Water Resource Consumption

Most of the energy required to operate AI models is consumed in data centers, and the carbon emissions generated in this process are substantial. Research indicates that the energy needed to generate 1,000 AI images is similar to the carbon emissions produced when driving a car approximately 6.6km.

Additionally, large amounts of water are used to cool data centers. Especially for data centers located in water-scarce regions, this can place a significant burden on the local community's water resources. According to Microsoft's research, a large data center can consume about 10 million gallons of water per day, roughly equivalent to the daily water usage of 15,000 households.

Resource Depletion Issues

Building AI infrastructure requires semiconductors, rare metals, and other special materials. Mining these resources is often linked to environmental destruction and can negatively impact ecosystems and local communities, particularly in developing countries.

Researchers warn that if the current AI boom continues, demand for these resources will surge, exacerbating supply chain problems and environmental impacts. The more trends like Ghibli-style image generation repeat, the more these problems are likely to worsen.

4. Solutions and Outlook: Finding a Better Balance

Approaches to Copyright Issues

Several approaches are being explored to address copyright issues with AI-generated content:

• Transparent Learning Data Disclosure: AI companies should transparently disclose the sources of data used for model training and obtain appropriate permission from rights holders for copyrighted works.

• Opt-out Mechanisms: Systems need to be built allowing artists to choose not to have their works used for AI learning.

• Compensation Systems: Arguments have been made that if an AI model learns and uses a specific artist's style, appropriate compensation should be provided to that artist.

• New Legal Frameworks: Discussions are ongoing about the need for new legal frameworks, recognizing that existing copyright laws are not suitable for the AI era.

Methods to Minimize Environmental Impact

Approaches to reduce the environmental impact of AI:

• Renewable Energy Use: AI companies, including OpenAI, should be encouraged to use renewable energy for data center operations.

• Model Optimization: Efforts are needed to develop more efficient algorithms and lightweight models to reduce the computational requirements for the same tasks.

• Carbon Offset Programs: Proposals include calculating carbon emissions from AI use and investing in carbon offset projects accordingly.

• Raising User Awareness: It is also important to educate users about the environmental costs of AI image generation and encourage them to use this feature only when necessary.

Ethical Balance in AI Technology Development

As AI models like GPT-4o continue to evolve, the Ghibli-style image generation feature is just one example of the creative possibilities these technologies offer. However, for such developments to be truly valuable, the following balances are necessary:

• Balance Between Innovation and Respect: Pursuing technological innovation while respecting the copyright and artistic identity of original creators.

• Balance Between Accessibility and Responsibility: Making AI technology accessible to more people while considering the associated social and environmental responsibilities.

• Balance Between Short-term Pleasure and Long-term Sustainability: Seeking a long-term sustainable direction for AI development rather than chasing temporary trends and fads.

5. Conclusion: Towards a Future that Recognizes Invisible Costs

The simple prompt "Draw in Ghibli style" vividly demonstrates the dual nature of AI technology. On one hand, it showcases remarkable technological advancement and creative possibilities; on the other, it casts shadows of copyright infringement concerns and hidden environmental costs. True technological progress must be accompanied by consideration not just of 'what can be done' but 'what should be done.' The trend of GPT-4o's Ghibli-style image generation provides our society with an opportunity to think more deeply about the costs and responsibilities hidden behind the beautiful results that AI creates.

In the future, we need a balanced approach that respects technological innovation and creativity while simultaneously ensuring artists' rights and environmental sustainability. In the process of finding that balance, it is important for each of us as users to become responsible consumers of AI technology. We need an attitude that enjoys the beauty of magical AI images but recognizes the invisible costs behind them and strives to make better choices. This is the moment to adopt such an approach.