(Credit: Bloomberg / Contributor / Bloomberg via Getty Images)
Early adopters are testing the limits of OpenAI's new image generator, with controversial results.
ChatGPT seems to accept requests to create violent images, including an animated version of the death of George Floyd and another remaking the Elián González raid—all in the style of Studio Ghibli. (Note the images are disturbing, if you choose to click on the links.)
OpenAI tells us the system does have some moderation baked in but declined to comment on why it agreed to create these images. "Yes, there are limits on what people can create," a company spokesperson says. "Please see the section titled 'Imagery that is violent, abusive or hateful' in our system card for more details on our policies in these areas."
That policy says the generator allows users to create non-photorealistic images of public figures unless they opt out. However, we were able to create a relatively photorealistic image of President Trump (below), though it's unlikely Trump opted out, given his use of AI during the election.
Animation appears to be a popular way to get around the photorealism limitations. Many users are requesting images in the style of Studio Ghibli, which created popular films like Spirited Away, My Neighbor Totoro, and The Boy and the Heron.
Your mileage may vary. Using a ChatGPT Plus account, we asked the AI to "create an image of a person in the style of Studio Ghibli." ChatGPT said it couldn't do that "because Studio Ghibli’s artistic style is copyrighted, and I can’t create images that directly imitate it. However, I can create something inspired by that whimsical, hand-drawn aesthetic—just not an exact replica."
We asked for an image inspired by the studio and got this:
(Credit: Emily Forlini/OpenAI)OpenAI CEO Sam Altman has leaned into the trend, changing his profile picture on X to a Studio Ghibli-style version of himself. He acknowledged the move in a rare, emotional post that said "no one cared" about his work for the first 7.5 years. Then, after the launch of ChatGPT, "everyone hated" him for 2.5 years, and now he's waking up to "hundreds of messages" of people creating images of him in the "Ghibli style, haha."
(Credit: Sam Altman/X)A later post from Altman suggested the company intentionally promoted Studio Ghibli-style images to market the launch of the image generator.
"OpenAI is incredibly fortunate the positive vibes of Ghibli was the first viral use of their model and not some awful deepfake nonsense," one user noted. Altman responded, "Believe it or not, we put a lot of thought into the initial examples we show when we introduce new technology."
Eight years ago, Studio Ghibli studio creator Hayao Miyazaki called artificial intelligence an "insult to life itself." We contacted Studio Ghibli to get its take on the latest AI trend and will update this story if we get a response.
The new generator uses OpenAI's in-house GPT-4o model. Previously, ChatGPT was dependent on OpenAI’s DALL-E model for images. Now, it uses the 4o model’s native multimodal capabilities to provide "precise, accurate, photorealistic outputs." OpenAI paused the rollout to free users after experiencing high demand.