Studio Ghibli-Based Imagery Proves AI Art Keeps Generating Our Gag Reflex

Forbes

28-03-2025

HOLLYWOOD - JULY 27: Director Hayao Miyazaki arrives to the special screening of "Ponyo" held at ... More the El Capitan Theatre on July 27, 2009 in Hollywood, California. (Photo by Michael Tran/FilmMagic)
This week, AI-generated imagery in the style of beloved Japanese animation production company Ghibli Studio began flooding the internet courtesy of a new filter available on the latest version of Open AI's GPT-40. The filter turns images like photographs into a simulation of the hand-drawn, pastel toned, cel-shaded look seen in Ghibli films such as Princess Mononoke, Howl's Moving Castle and The Boy and the Heron.
While some people are beguiled by the tool, the reaction that has greeted this surge of faux anime art is the same that we've seen over and over again when AI artwork breaks into the popular consciousness: disgust at both the aesthetics and the ethics. This time, the stakes are amplified because the project aims to diminish criticism of AI art by diminishing one of its most prominent critics.
For examples of previous negative reactions, consider the example of the Coca Cola commercial released this holiday season, in which hundreds of hours of AI-generated video footage was edited down to a fast-cut 30 second spot designed to play on sentimental remembrances of a classic holiday ad from the 1990s. Though technically impressive in some ways, even squeezing professional level AI tools as hard as possible still produced a few uncanny artifacts in details and scale. Most of them went by quickly enough to evade conscious perception, but the overall effect was unsettling. Online critics highlighted the imperfections using frame grabs and zooms, raising issues that ordinary human-created, computer-assisted 3D animation, by its nature, rarely exhibits.
Toys-R-Us, another brand with nostalgic connections to childhood, got caught in the AI honeytrap last summer with an ad inspired one commenter to write on X, 'There's something so cynical about creating an ad about the limitless imagination of a child and then rendering it with soulless AI slop.'
Sometimes these campaigns include elements meant to anticipate criticism of AI and turn that into part of the story. Liquor brand Absolut Vodka launched a bizarre campaign on YouTube aimed at promoting acceptance of AI imagery but suggesting that we be more inclusive about the people and ethnicities depicted in synthetic photographs, so that future AI models scraped from these images would themselves be more diverse. This did not have the intended effect, and got the campaign listed among 'The Bad' implementations of AI in advertising by Creative Bloq.
AI imagery is particularly loathed in the world of comics and graphic novels. DC got embroiled in a controversy last year when a fellow artist spotted telltale AI imperfections in the artwork for a cover of Superman by artist Francesco Mattina, resulting in the cancelation of solicitations for the edition of the comic featuring that cover and a strong statement from DC and other comic book companies clarifying their policies and prohibitions regarding its use.
In all of these cases, the negative response to the AI imagery occurs on several levels. Our brains are aware of inconsistencies in the imagery at a detailed level, even if the exact nature of these inconsistencies eludes our conscious perception. This is similar to the 'uncanny valley' we experience when we see computer-generated imagery, usually realistic 3D animation, intended to duplicate reality but missing some of the subtle atmospherics of lighting and physics that we'd see in nature. At least with CGI, the shortcomings are consistent: things like shadows and mist always don't look right, or objects in the distance appear a little too clearly. Over time, we can identify the things that 'look wrong,' and they reflect lapses in taste or craft by humans using the technology as much as limitations of the technology itself.
With AI, the artifacts occur at random because of how the diffusion models generate imagery. In still images, everything can look 'right' except that the pattern on the drapes doesn't follow any consistent design logic. Or a piece of machinery is rendered perfectly in terms of materiality, reflection and mass, but the machine includes gears or controls that have no clear function. In video imagery, small changes frame to frame can lead to jitter in the background or problems like, in the Coke ad, people in the crowd looking too big relative to the scale of other objects in the scene.
In depictions of humans, the effect is even more pronounced. According to emerging research, uncanny valley reactions sometimes occur when we pick up on mismatched features, like realistic eyes but unrealistic skin. This particular incongruity is a common problem for stable diffusion-generated AI images.
Even if the people responsible for prompting the AI systems notice these problems, there isn't much they can do about them. Post-processing AI-generated images to remove these subtle issues is almost as time-consuming as creating the imagery from scratch, and requires the kind of technical mastery that the AI systems are designed to replace.
Typically, the MO of the proponents of AI video and imagery is to suggest that we lower our standards to meet the capabilities of the system, recognizing the technical virtuosity of getting things close enough, especially considering the savings in time and effort.
That on its own might be enough to satisfy some people, especially since emerging tech like the Ghibli simulator are getting pretty good at avoiding the 'extra finger' problems that bedeviled early generations of AI imagery. But then there are the overarching ethical and economic considerations. As most people know by now, online critics have the knives out for AI imagery because the models were built with imagery obtained without consent, compensation or control, while threatening the livelihoods of those it has stolen from. This is picking up steam as more people, not just artists, begin to realize that AI is coming for their jobs as well.
AI proponents, both in business and on the consumer side, are obviously tired of this criticism. They are invested, financially or emotionally, in the ultimate success of this technology and they don't want to hear any more whining from a bunch of artists and elitist critics. Billions, maybe trillions, of dollars are on the line, and companies are not going to let a little consumer disgust and rejection get in the way of their big plans. After all, they're the masters of the universe, and we are their increasingly disempowered 'end users.'
That edge of 'what are you going to do about it?' gives the Ghibli situation added cultural salience. Hayeo Miyazaki, Ghibli's founder and artistic visionary, famously objected to AI art in the strongest possible way, saying he found it disgusting and anti-human. Building an AI art tool based on his aesthetic is basically spitting in the face of a global icon whose work has brought joy to millions, but not only that. It's a way for the AI tech lords to show all the aggrieved dissidents who's boss, as we do these days.
In a sense, this seems to set up a final battle between the forces of 'wow, this is really cool!' and the forces of 'ick, stop, this is gross on every level!' employing the charming Ghibli aesthetic to divide the opposition. After all, who among us can resist turning ourselves into anime characters? And once we're all implicated, then maybe we'll shut up and let them make money.