Creation and AI: AOKI Studio Interviewed by the Nikkei

Questions

Since when has AOKIstudio been interested in AI?

We were among the first testers of MidJourney in June 2022, before the official announcement in July 2022.
We then tested and compared DALL·E and Stable Diffusion simultaneously.
In March 2023, we discovered Adobe Firefly, automatically integrated into Photoshop.

At first, like many others, we were amazed by the apparent quality and originality of the generated images.
We did not yet know how, but it felt like a revolution that would transform the creative fields.
We specifically studied MidJourney, whose users' prompts are public, unlike DALL·E and Stable Diffusion, whose prompts are private.
Even today, with MidJourney, we have the possibility of analyzing the prompts behind the generated images and thus understanding how these systems work.

Three points caught our attention:
From the very first days, we were surprised to discover that tagging artists' names in prompts was a common practice: in the manner of..., in the style of...
This included artists whose works are in the public domain, like Leonardo da Vinci or Van Gogh, but also very often more recent artists whose works are supposed to be protected by copyright.
The list is endless, but among the most tagged, we find artists such as Takashi Murakami, Yayoi Kusama, production companies like DreamWorks and Pixar, film directors like Tim Burton and Wes Anderson, licensed characters like Pokémon or the Minions, designers and architects like Tadao Ando and Zaha Hadid, fashion designers like Vivienne Westwood, and graphic designers like Paul Rand and Milton Glaser.

We also discovered that the same prompt could generate extremely different images and that the generated images rarely matched what we had in mind.
The apparent originality and quality of the results often led us to select an image by default, and the practice consisted of generating a very large quantity of images to select just one.

The third surprising point was that the users of AI tools were not the artists from before AI, but rather people who had never imagined producing images before.
AI allowed everyone to generate an avatar for social media, a logo for a company, an object or space design for a project proposal, or simply to create images playfully, just for fun.

What are the current key issues regarding AI?

The main issues emerged as early as 2022: copyright, the disappearance of certain professions, cognitive decline, creator-consumer ambiguity, overproduction, and saturation.

A more recent issue is that of pollution: pollution of our planet, with the growing consumption of electricity and water for server cooling, and pollution of our imagination, with an exponential growth in the production of images.
The trend is also moving, however, towards a multiplication of tools, with already more than thirty AI image generators by early 2025.
Some new features seem to reinforce the initial practices, namely generating a very large number of images to select just one.
In 2022, MidJourney generated an image in 30 to 60 seconds.
In 2025, Leonardo.ai’s new tool, Flow State, offers a real-time, continuous, and infinite stream of images based on a single prompt.

Two years after the emergence of AI tools, it seems that these systems are beginning to feed on their own production, and the feeling that results are increasingly tending toward an average is particularly relevant today.

The tools also tend to specialize and integrate the ability to rework an existing drawing or photograph.
Will these tools leave less room for the randomness of a simple prompt and open the way to greater creativity?

What are the dangers of creators becoming dependent on generative AI?

Every creative person,designer, illustrator, musician, writer, cook, and more recently, computer coder — learns and perfects their skills throughout their life.

The creative act is, in itself, an act of learning that does not simply consist in reproducing a recipe identically, but rather in continuously questioning oneself, doubting, testing new methods, making mistakes, going back, and eventually surprising and astonishing oneself.
If the creative act were not an act of learning, we would simply always do the same thing.
It is because we learn through creating that we are able to create new things.

Without creative action, there is no learning, and without learning, no new creations — and this is precisely what is happening with the use of AI image generators.
The operating mode of programs like MidJourney or DALL·E erases the learning dimension of the creative process.
The result appears without the need to go through the traditional stages of creation.

Dependence on AI is obvious; many people already admit they can no longer do without it.
The main danger of this dependence, beyond the misleading impression of creating something oneself, is therefore the loss of learning.

Traditional creative work consists of giving, whereas generating a design or a visual with AI consists of receiving.
Dependence on AI can then transform us, the producers, into consumers.
We no longer create; we consume.
In exchange for a paid subscription to an AI system, we place orders until we are satisfied.

Why is learning through imitation — natural for humans — controversial for AI?

It is necessary to use large amounts of data to develop algorithms based on deep learning, and this is not a problem in itself.
The issue arises from the fact that private companies have used copyrighted works for commercial purposes without the authors’ permission.
Put yourself in the shoes of an artist who has created a character for their future manga. They’re not exactly thrilled at the idea of their character being plagiarized and then commercially exploited by others, right?

This is not only about AI learning, but also about how it is used.
Don’t we see that when searching for a drawing of the character Totoro by Hayao Miyazaki on the Internet, we no longer find the original illustrations, but only distorted imitations of the character?

Can we still speak of human creation when 90% of the result comes from AI?

It is indeed no longer possible to clearly distinguish what comes from human contribution and what has been generated by AI.
We may be entering an era where this question is losing its meaning. No matter how something was produced, everything becomes potentially true or false, and it no longer matters whether a human being or a machine is behind what we consume.

Does the perception of AI by creators vary between countries?

No, at first glance, the practices seem quite similar.
In any case, the tools being used still mostly come from the USA.