OpenAI eases rules on ChatGPT image generation as Ghibli AI trend goes viral | Technology News
Amid the social media frenzy over ChatGPT-generated Ghibli-style art, OpenAI has loosened its content moderation policies to allow its AI model to generate images depicting public figures, hateful symbols, and racial features.
Until now, ChatGPT has rejected prompts users to generate such images on the basis that they were too controversial or harmful. However, the Microsoft-backed AI startup’s updated rules now permit ChatGPT to generate and modify images of public figures such as US President Donald Trump, tech billionaire Elon Musk, and others.
Elaborating on how the company’s approach has “evolved”, Joanne Jang, the head of product and model behaviour at OpenAI, said, “We’re shifting from blanket refusals in sensitive areas to a more precise approach focused on preventing real-world harm. The goal is to embrace humility: recognizing how much we don’t know, and positioning ourselves to adapt as we learn.”
Story continues below this ad
“AI lab employees should not be the arbiters of what people should and shouldn’t be allowed to create,” read the blog post Jang published Friday, March 28. She also pointed out that OpenAI has decided to establish an opt-out l for public figures who do not want to be depicted ChatGPT.
In a separate white paper released last week, OpenAI said it will allow ChatGPT users to “generate hateful symbols,” such as swastikas, in educational or neutral contexts, as long as they don’t “clearly praise or endorse extrem agendas.”
Jang also said OpenAI has revised its definition of offensive content, suggesting that ChatGPT will now accept user prompts around physical characterics such as “make this person’s eyes look more Asian” or “make this person heavier”. The previous guidelines unintentionally implied these attributes were inherently offensive, she added.
Last week, OpenAI launched native image generation in ChatGPT through GPT-4o, sparking a viral trend of users requesting Studio Ghibli-style renderings of their selfies, wedding pictures, family portraits, and more.Story continues below this ad
The surprisingly flexible guidelines for the ‘Images with ChatGPT’ feature reportedly state that “the model can generate images that resemble the aesthetics of some arts’ work when their name is used in the prompt.” It is also reportedly allowed to mimic the styles of creative studios such as Pixar or Studio Ghibli, despite the brewing debate over training AI models on copyrighted works and the fair use exception.
OpenAI’s changes to its content moderation rules also align with its broader push to “uncensor” ChatGPT, with the culture wars over AI safety intensifying since Trump’s return to the White House.
© IE Online Media Services Pvt Ltd
Expand
