As you might have already seen, Ghibli-styled image generation (or conversion) has pretty much shaken the world. Soft-toned Ghibli pictures have hooked the internet, but I am not a fan, which I will discuss later in the article. But first, let's learn what OpenAI’s latest GPT-4o image generation is really about.

What OpenAI’s GPT-4o image generation can do

OpenAI’s new GPT-4o model brings a range of improvements to AI-generated images, making them more interactive, flexible, and detailed than ever before. Some key upgrades include:

• Text Rendering: One of the biggest changes. GPT-4o can now generate legible text inside images. No more warped, gibberish letters that look like a bad dream. This means AI-generated signs, labels, and posters are now readable.

• Image Modification: Instead of starting from scratch each time, GPT-4o allows users to tweak an image dynamically through simple text instructions. You can add elements, remove them, or refine details without needing a new prompt.

• Transparent Backgrounds: A new feature that lets AI generate images with no background, making them easier to use in design projects.

• Multi-Object Handling: GPT-4o is better at managing complex images with 10 to 20 objects, whereas older models would struggle beyond 5 to 8 objects.

• Blending Images: You can give it multiple pictures, and it can merge them effortlessly while maintaining style consistency.

Improvements and Limitations

The results are undeniably better than before. The interactive nature of editing AI-generated images makes the process more intuitive, and the ability to generate readable text inside an image is a much-needed improvement. However, OpenAI admits that issues still exist:

• Cropping Issues: Some images are cut off at the bottom for no apparent reason.

• Hallucinations: AI still generates weird, nonsensical details if the prompt is too complex.

• Non-Latin Text Struggles: While English text is clearer, other scripts like Chinese, Japanese, and Arabic still get distorted.

• Too Many Objects? It manages 10 to 20 objects well, but anything beyond that can break the coherence.

The Ghibli problem: Why AI-generated "art" feels hollow

Many have noticed that OpenAI and other large-scale corporations use data illegally to train their models, including for image generation. The same goes for Ghibli-style AI art. This, of course, highlights how big corporations get away with theft while Aaron Swartz lost his life for far less of a "crime."

But this isn’t even the real issue. What personally bothers me stems from what Walter Benjamin writes in his essay The Work of Art in the Age of Mechanical Reproduction. For art to be meaningful, it’s not just about skill, aesthetics, or even the "cultic status" of the artist (the idea that something is valuable just because Picasso made it). As Benjamin puts it:

“...even the most perfect reproduction of a work of art is lacking in one element: its presence in time and space, its unique existence at the place where it happens to be.”

Hence, all in all, art isn’t just about an output. It’s a human condition both personally experienced and socially mediated. It’s the artist’s lived moment, their perception, their ability to intuitively feel and reflect their emotions into something tangible. But this also isn't really an endorsement of the essentialist conception of art, that art in itself contains an inherent essence, atomized from the historical specificity and understood as a sense-immediate product with private aesthetic experience, but rather Art is something mediated by social relations and the reflections of its conditions.

AI can mimic styles, but it cannot experience them. To be clear, it's limitation is not just a lack of "human experience" but an absolute absence of historical consciousness. Art is created through human labor, whereas AI scrapes, synthesizes, and commodifies and ultimately reducing art to mere decoration rather than a socially embedded force. However, this is not a romantic critique that insists art must require human hands. History reshapes production, and in a future where all labor might be subsumed under automation, even art could become an automated process. But even then, AI does not generate art autonomously; it functions as a tool built not upon human organs but alien to humans themselves. The issue is not simply that AI "lacks emotion" but that it operates outside the process of dialectically embodied cognition. Its mode of production is foreign to the human process that has always defined artistic creation.

Does Miyazaki hate AI art?

AI-generated Ghibli images, for example, might be aesthetically pleasing to some, though they don’t work for me. To me, they seem like nothing more than glorified Snapchat filters trying to apply a soft tone. The mass production of these images, at any given moment, inevitably loses its "aura," as Benjamin would put it. However, when we look at perfect mimics of images, they lack something crucial... a mediated experience that can only come from human consciousness.

This reminds me of a particular video featuring Hayao Miyazaki, where engineers demonstrated an AI-generated video to him. He remarked that he felt disgusted, calling it "an insult to life itself." Miyazaki, who famously said that a particular 4-second clip in his films required 15 months of painstaking work, emphasized that it was all worth it in the end. The production of such Ghibli-like images, when used to showcase the mimicry of violent and grotesque scenes from both historical and present times, reflects the bizarre world we live in.

Miyazaki, a person who has long championed environmentalism and love, would have to witness the creation of work that is so mechanically produced, where entire global energy resources, including deforestation, were exploited. This is where art and human connection come into play: the authenticity is ultimately alienated from both producers of art and those with sensual capabilities.

Intuition or Mechanical Calculation?

There was a time when a massive question arose between the genius of Einstein and von Neumann. The question was: who was the true genius? On one hand, both had made significant contributions to the field of science. But it was von Neumann himself who declared that the extraordinary thing about Einstein, which he lacked, was intuition. Einstein was not merely a human calculator but an original thinker who thought intuitively, not mechanically.

Although this is a different domain from what we’re discussing, von Neumann's contribution is undeniably significant. However, when it comes to the development of human consciousness, intuition plays a crucial role in shaping the trajectory of the world. It’s through intuition that humans use history as a means to achieve their aims. While the question of art can’t be reduced to just a specific domain of intuition or creativity, it instead reflects the process of world-building itself. The same applies to all AI-generated art. So ultimately, what i mean is, instead of condemning AI 'art' as bad, we must analyse the reason it feels so hollow, not in moral terms of course, but in its social form, through which it stagnates the process of developing and understanding our own humanity.

Final thoughts

GPT-4o’s image generation is impressive. It’s a significant leap in AI art, especially with better text rendering, layering, and object handling. But at the end of the day, no matter how advanced AI becomes, it can’t replicate the human experience behind art. AI can generate Ghibli-styled images, but they will never be Ghibli.