In April 2026, Hayao Miyazaki's old quote — "an insult to life itself" — re-circulated for the dozenth time, this time in response to OpenAI's GPT-4o-driven Ghibli-fication trend. I want to talk about a quieter version of the same theft. Not the visual one. The naming one.
Throughout the spring of 2026, conversations about AI's appropriation of Studio Ghibli's visual aesthetic intensified. Goro Miyazaki, Hayao's son and director in his own right, gave interviews to The Hollywood Reporter and other outlets renewing the family's objections to AI-driven imitation of his father's work. The Miyazaki quote — originally delivered in a 2016 documentary about an early-stage AI animation demo, but evergreen — circulated again on every social platform with renewed weight. The visual theft is the headline. The naming theft, which I have been watching for two years, is the footnote.
The naming theft is real. ChatGPT, Gemini, and Claude have, across systematic prompt audits, converged on a remarkably consistent set of "Ghibli-coded" names when asked for soft-fantasy or whimsical baby names. The names are vowel-heavy, Japanese-leaning or Japanese-adjacent, and lifted from a corpus that includes Studio Ghibli scripts, fan wikis, and adjacent fantasy fiction. Parents using AI for naming inspiration are getting a Miyazaki style guide they did not consciously request.
The Ghibli Name Cluster
The cluster is identifiable. It includes Kiki (Kiki's Delivery Service), Sora (a recurring soft-syllable name in Japanese animation broadly), Yuna (also from gaming and anime), Lumi (lifted from various fantasy contexts), Kai (Studio Ghibli-adjacent and now broadly anime-coded), Hina, Sakura, Mei (My Neighbor Totoro), and a few invented names that scan as Ghibli-coded without specifically matching any film character. The unifying features are: short, vowel-heavy, soft-consonant, two-syllable, evocative without being heavily symbolic.
If you ask any of the major large language models for a soft fantasy girl name, you will get one of these names with high probability. If you ask for ten, the list will be dominated by them. The model has, through whatever combination of training corpus and reinforcement signals, decided that the Ghibli aesthetic is the canonical answer to the soft-fantasy-name prompt.
This is not a tiny phenomenon. AI-driven naming inspiration has become, in 2026, one of the most-used tools among expectant parents. Every major naming-focused publication has acknowledged it. ChatGPT's parenting-related queries have been growing year-over-year at rates that exceed almost every other consumer use case. Parents are asking. The model is answering. The answers, when soft-fantasy is the request, are Ghibli-coded.
What This Has To Do With Theft
The visual Ghibli-fication trend that drew Goro Miyazaki's renewed objections is a clear case of AI training on copyrighted artistic work and producing imitations of it without permission, attribution, or compensation. The studio's claim is straightforward: they made the aesthetic, they own the aesthetic, and AI's appropriation of it is a violation that the law has not yet figured out how to redress.
The naming version of the same problem is fuzzier but structurally similar. Names themselves are not copyrightable. The character Kiki cannot be trademarked as a baby name in any meaningful sense. But the model's tendency to suggest Kiki specifically — over Helena, Eleanor, Iris, Hazel, or any of dozens of equally good fantasy-soft girl names — is a function of training on Ghibli's narrative output. The model knows Kiki because Kiki appears in transcripts and fan wikis and reviews of Kiki's Delivery Service. The model recommends Kiki because the aggregate of those sources weighted Kiki as a soft, vowel-friendly, fantasy-coded name.
The studio's contribution to the model's recommendation is real. The studio gets no acknowledgment, no compensation, no opt-out. Parents using the recommendation generally do not know that the suggestion is, in part, a Studio Ghibli style guide rendered through statistical inference. They think they are getting an objective recommendation from a neutral assistant. They are getting, in important measure, a watered-down Miyazaki.
What Hayao Miyazaki Would Say
Miyazaki has been clear, across multiple decades of interviews, that he resents the appropriation of Studio Ghibli's distinctive sensibility for purposes the studio did not authorize. The 2016 quote — "an insult to life itself" — was delivered specifically about an AI-driven animation demo. Miyazaki's objection was not narrowly technical. It was philosophical. He believed that AI-generated approximations of his work flatten the human creative effort that produced the originals, and that the flattening is a kind of disrespect to the labor and to the people who produced it.
The naming version of the same objection is, I believe, one Miyazaki would recognize. The model is taking the names his films created or popularized — names that emerged from specific human creative labor in writing rooms, animation cels, voice acting — and offering them to American parents as friendly, frictionless suggestions. The labor is invisible. The provenance is unmarked. The names are presented as if they are universal properties of the linguistic landscape rather than specific products of a specific studio's creative tradition.
Miyazaki has not, to my knowledge, commented specifically on the AI-name-recommendation phenomenon. I doubt he will. He is too senior and too generally disinterested in the surface forms of the AI debate. But the naming version of the appropriation is, in structure, the same thing the visual version is. The model is monetizing what the studio built, without the studio's permission, in a way the studio cannot stop.
What Parents Are Actually Adopting
Here is the awkward part. American parents have not, at least so far, adopted the Ghibli-coded name cluster at the scale ChatGPT recommends it. Kiki is not a Top 1,000 name. Lumi is not a Top 1,000 name. Yuna is at the long tail. Mei has held steady but not surged. The Ghibli-coded names are the model's favorites and parents' polite declines.
The names that are actually growing fastest in the SSA chart's soft-fantasy adjacent zone are different. They are Aurora (Disney, not Ghibli). They are Luna (Spanish-English crossover, only Ghibli-adjacent). They are Iris and Wren and Eleanor and the broader vintage-revival cluster. Parents who consult AI for naming get Ghibli-coded suggestions, then go their own way. The model's recommendations are sounding boards, not commitments — a phenomenon I covered in detail in our Elara piece earlier this week.
The result is a strange compromise. The model is appropriating Studio Ghibli's naming aesthetic at scale. Parents are not actually using the appropriation. The appropriation is happening at the recommendation layer without showing up in the adoption layer. The studio is, in a quiet way, being raided without being meaningfully amplified. The theft is real. The benefit is, mostly, absent.
The Counter-Reading
The honest counter-case is that AI's recommendation of Ghibli-coded names is no different from a magazine writer's recommendation of them. Naming guides have been pulling examples from Ghibli films for years without complaint. The studio has not, to date, objected to its character names being shared in baby-name lists. The model is just doing, at scale, what naming editors have done in print. The novelty is in the scale, not in the act.
I think this is a fair counter, but it under-weights the difference between a magazine recommendation (which generally cites the source — "Kiki, after Kiki's Delivery Service") and a model recommendation (which generally does not). The provenance asymmetry matters. The model launders the source. The magazine does not.
It is also fair to point out that some Ghibli-coded names are genuinely Japanese baby names that exist independently of Studio Ghibli's films. Sakura is a real Japanese name with no Ghibli specific origin. Sora, Yuna, and Mei are similarly broader-than-Ghibli. The model recommending these is recommending a broader cultural pool that Ghibli films are part of but do not own. The line between Ghibli appropriation and broader Japanese-naming-pool engagement is not as clean as my framing suggests.
What I Think Should Happen
I do not think Studio Ghibli is going to sue OpenAI over baby-name recommendations. That is not a real legal pathway. What I think should happen is that the AI naming-recommendation infrastructure should learn to attribute. "Kiki" recommended for a girl should come with a note: this name is associated with Kiki's Delivery Service, a Studio Ghibli film. Parents should know what they are picking. They should know whose imagination they are reaching into. The recommendation should be honest about its sources.
This is not a difficult feature to build. It is one prompt-engineering pass away. It is the kind of thing that would, if implemented, restore some of the dignity the visual Ghibli-fication trend has eroded. The names came from somewhere. The somewhere is named. The names of the somewhere should travel with the recommendations.
The Closing Insult
Miyazaki's "insult to life itself" quote was about AI flattening the creative labor that produced his films. The naming version of the same flattening is quieter and probably permanent. AI will continue to recommend Kiki and Sora and Lumi to parents. Parents will continue to mostly not adopt them. The studio will continue to receive no acknowledgment. The aesthetic debt will compound.
The strangest part is that the names themselves are good names. Kiki is a charming name. Sora is a charming name. Lumi is a charming name. Studio Ghibli built a small, beautiful naming canon over four decades, and the canon is genuinely contributory to the broader vocabulary of soft-fantasy naming. The work was real. The names were the products of real care. They deserve to be remembered as such, not laundered as model outputs. If you choose Kiki for your daughter, you should know you are choosing a name from Hayao Miyazaki's imagination. That is a beautiful sentence. It is also a sentence the model will not write for you. So I am writing it for you here.
Data source: U.S. Social Security Administration. Analysis by NamesPop.
Found this helpful?
Share it with someone who’s picking a name.