Our story opens on March 25 — mere weeks ago, yet somehow already fading into the fog of our eternal present — with OpenAI releasing a new version of ChatGPT. Compared with previous versions, this one was surprisingly good at generating novel digital images: Users could tell it what they wanted to see, and voilà, an image would appear. And what lots of people wanted to see, it transpired, was real photographs transformed to look like stills from the animated films of the Japanese company Studio Ghibli.
These beloved movies — especially those directed by Hayao Miyazaki — evoke the wonder and innocence of childhood but also the forces that erode that innocence: mortality, history, greed, hubris. They are products of a labor-intensive and largely by-hand animation process; nothing else looks quite like them. It’s not surprising that people were tickled by the prospect of near-instantly making anything look like a Ghibli movie, complete with that trademark aura of cozy but sophisticated wholesomeness. But ChatGPT’s Ghiblification set what was surely a new speed record for the emergence of a meme format: Within 24 hours of the new release, generating Ghibli images had become the way for people to demonstrate their fluency in the internet’s shifting codes.
On the afternoon of the 25th, an engineer named Grant Slatton racked up tens of thousands of likes on X after posting a Ghiblified photograph of himself, his wife and their corgi on a beach. Soon people were posting not just family photos but images from the news and from history: a Ghiblified Donald Trump; a Ghiblified Jeffrey Epstein; Ghiblified jets smashing into Ghiblified twin towers; a Ghiblified murder of George Floyd. They started Ghiblifying old memes: Ghiblified “distracted boyfriend,” Ghiblified “Bernie Sanders at Trump’s inauguration,” Ghiblified “Ben Affleck mournfully smoking a cigarette.” Refreshing social media, you could watch in real time as internet culture rummaged around its cobwebbed pantry, tossing everything into its new game. Sam Altman, chief executive of OpenAI,claimed that a million new ChatGPT users signed up in a single hour on March 31 and that the Ghibli-related demands were “melting” the company’s graphic processors.
Advertisement
SKIP ADVERTISEMENT
Haters like me felt obligated to point out that this Ghiblification seemed likely to rely on ChatGPT’s having been fed Ghibli movies as training, with no permission sought or compensation offered. (All OpenAI has said on the subject is that the new model was trained on “images reflecting a vast variety of image styles.”) These movies are painstakingly constructed stories about the irreducibility of the human spirit and the fragile beauty of nature. The Ghiblified images are something else, something that wouldn’t be out of place in a Miyazaki movie: a swarm of cheap knockoffs feeding parasitically off the essence of the originals, cranked out by a technology so plunderously energy-intensive that coal plants slated for closure have been kept open just to keep it running. But pointing this out risked sounding like a killjoy: They were just memes, right?
By March 27, the meme had reached the White House, or at least its official X account, where a news release about the planned deportation of a Dominican woman — a convicted fentanyl dealer — was paired with a Ghiblified image of this woman weeping in shackles.
Subscribe to The Times to read as many articles as you like.
