ChatGPT’s Studio Ghibli-Style Images and the Rise of Copyright Concerns

Recently, social media has been alive with images that look like they belong in a Studio Ghibli film. Users have transformed their selfies, family photos, and memes into works that showcase the soft pastel colors and unique visual style that Studio Ghibli, founded by Hayao Miyazaki, is famous for.

This trend follows a significant update to ChatGPT by OpenAI, which enhanced its image generation abilities. As a result, users can now create stunning Ghibli-style images in just seconds. The feature's popularity surged to the point that the system occasionally crashed from overwhelming demand.

Generative artificial intelligence systems, such as ChatGPT, can be seen as 'style engines.' They enable users to craft images by offering more precision and control than ever before.

How the New ChatGPT Generates Images

Generative AI programs create outputs based on user prompts, including requests for images. Earlier AI image generators relied on diffusion models that slowly refined random, noisy data into a clear image. In contrast, the latest ChatGPT utilizes an ‘autoregressive algorithm’ that treats images similarly to language.

This algorithm breaks images down into ‘tokens,’ allowing ChatGPT to predict various visual components just as it might predict words in a sentence. Thanks to tokenization, the AI can accurately create images based on specific user prompts, enabling users to modify particular features while maintaining the integrity of the overall image. This also addresses the long-standing problem of generating correct text within images.

A unique strength of generating images within a large language model is the ability to reference the extensive knowledge encoded within the system. Therefore, when users mention Studio Ghibli, the AI readily understands the reference without needing detailed descriptions.

The recent Ghibli trend started with OpenAI and quickly gained traction among Silicon Valley engineers. Various groups, including government officials, have leveraged this technology for creative purposes, leading to unexpected applications, such as the White House producing a Ghibli-style image related to immigration issues and the Indian government promoting its vision of a 'New India.'

Recognizing AI as 'Style Engines'

Generative AI systems do not store information like traditional databases. Instead, they encode various elements—text, facts, image fragments—as patterns or 'styles' within their neural networks. These AI models, trained on vast amounts of data, learn to identify patterns at different levels. Lower layers might recognize basic features while higher layers capture more complex concepts or visual elements.

As a result, everything—including objects, styles, genres of writing, and professional tones—is transformed into recognizable patterns. When AI learns about Miyazaki’s work, it does not store specific Studio Ghibli frames. Instead, it encodes ‘Ghibli-ness’ as a mathematical style applicable to new images.

This process of encoding and transferring styles has become a primary objective in visual AI. The newly enhanced generator allows for unprecedented scalability and control. If all elements can be viewed as styles, users can mix and match these styles to create unique combinations.

The Copyright Controversy: When Styles Become Identities

While the ability to work with styles powers generative AI, it also raises valid concerns. Many artists find it troubling to see their distinct artistic expressions reduced to mere styles that anyone can replicate with simple text prompts.

Hayao Miyazaki has previously expressed criticism of AI but has not made a public comment on the ongoing trend of generating images in his renowned animation style. This situation brings about serious questions regarding copyright and creative ownership.

Traditionally, copyright law does not protect styles as it only covers specific artistic expressions. One cannot copyright a genre of music, like 'ska,' or an art movement, like impressionism. This limitation is crucial. If someone could monopolize an entire style, it would inhibit creativity for everyone else.

However, there’s a notable distinction between general styles and highly recognizable ones that are closely tied to an individual’s identity. For example, when AI generates work ‘in the style of Greg Rutkowski,’ a Polish artist whose name appeared in over 93,000 prompts in the AI image generator Stable Diffusion, this could threaten his career and artistic legacy.

Some creators have already initiated legal action. In late 2022, a group of artists launched a lawsuit against several AI companies, claiming their image generators were trained on original works without consent, now allowing for the creation of derivative works imitating their unique styles.

As technology advances outpaces the legal system, efforts are underway to draft new laws that balance the need for innovation with protecting artistic identities. Regardless of the outcome, these discussions highlight the transformative power of AI style engines, emphasizing the importance of recognizing both their creative potential and the need for more nuanced protections for distinctive artistic styles.

ChatGPT, Studio, Ghibli, AI, Copyright