OpenAI just made it harder to turn your pics into Studio Ghibli-style images

Cartoons of pets, families, and famous memes made to look like the cartoons from the famed Japanese animation house Studio Ghibli flooded social media this week.

The chatter around the generated images, which came from ChatGPT's image generation model released Tuesday, even merited a joke from OpenAI's CEO, Sam Altman.

But the cutesy images didn't last long.

By Wednesday night, the system started to block some users' requests for images in the style of Ghibli and some — though not all — other artists.

OpenAI said in an explanation about GPT-4o's latest image generation technology that it was taking a "conservative approach" to images using artists' work in this model.

"We added a refusal which triggers when a user attempts to generate an image in the style of a living artist," the company said.

An OpenAI spokesperson told Business Insider on Wednesday night that the company continued to prevent "generations in the style of individual living artists" but did permit "broader studio styles."

"We're always learning from real-world use and feedback, and we'll keep refining our policies as we go," the spokesperson added.

The internet became awash with Studio Ghibli-esque photos after ChatGPT upgraded its image generator on Tuesday.

The Japanese animation studio has long inspired fan art and memes, but the current wave of photos took on particular significance because the studio's creator has criticized AI.

When Hayao Miyazaki saw AI-generated art at a presentation in 2016, he said, "I strongly feel that this is an insult to life itself."

Studio Ghibli did not respond to a request for comment.

The photos also reignited a debate about the copyright lines for generative AI companies.

A cat photo goes Ghibli on different AI models

On Wednesday night, ChatGPT blocked me from turning a cat photo into a Studio Ghibli-style image.

The AI model said my prompt didn't follow its content policy.

But when I hopped over to Grok 3, things got more magical. With the same prompt, I received exactly what I'd sought: a cat pictured as if it were in the popular film "My Neighbor Totoro." It was adorable — even though it had an extra leg.

Claude 3.5 Haiku, on the other hand, wasn't so great at mimicking Ghibli. With the same photo and prompt, it gave me what it charitably calls an "artistic interpretation" that was supposed to channel that classic hand-drawn, whimsical feel.

It looks like a potato with a mole on it.

I tested two more artists to see whether the rules held. Next up: Jeff Koons' sculptures.

Related stories

Business Insider tells the innovative stories you want to know

Business Insider tells the innovative stories you want to know

ChatGPT budged, giving me a metallic, balloon-like version of the cat photo I had put in. It's very Koons-like, with shiny, reflective surfaces.

Grok followed suit with a more lifelike cat, missing the signature balloon feel that Koons is known for.

Claude stuck to copyright rules and gave me what looked like a snowman with its face turned upside down.

Finally, I tested Andy Warhol's pop art style using a photo of myself.

Once again, ChatGPT blocked my request, citing content policy. I'm not sure why it let the Koons request through, but not Warhol. (Maybe Koons' legacy just doesn't hit the same way?)

Grok delivered big time with a vibrant, neon-colored image of me in Japan during cherry blossom season, straight out of a Warhol pop art dream.

Claude produced a multipanel piece — without my image. Creative.

xAI and Anthropic did not respond to a request for comment, sent outside business hours.

Ali Barr contributed reporting.