OpenAI's new image generator in ChatGPT is drawing attention for more than viral Studio Ghibli-style pictures. The release also marks a notable change in how ChatGPT decides what kinds of images it can create, edit, or refuse.
The company is loosening some earlier restrictions around sensitive content. Under the updated approach, ChatGPT can now handle certain image requests involving public figures, hateful symbols, racial features, and broader physical characteristics, while still refusing many sensitive prompts.
What Changed In ChatGPT Image Creation
This week, OpenAI launched a new image generator in ChatGPT. The tool quickly spread online because users could create Studio Ghibli-style images, but the bigger product shift is broader than one visual trend.
GPT-4o's native image generator improves ChatGPT's ability to edit pictures, render text, and represent spatial relationships. Those upgrades make the tool more capable as a general image creation system, not only as a novelty generator.
At the same time, OpenAI has changed parts of its content moderation policy. Requests that ChatGPT previously rejected as too controversial or harmful can now be allowed in some cases.
The newly permitted areas include:
- Images depicting public figures.
- Images involving hateful symbols in educational or neutral contexts.
- Image edits or generations involving racial features.
- Requests tied to physical characteristics such as eyes or body size.
OpenAI's model behavior lead, Joanne Jang, described the policy direction in a blog post published Thursday. She said, “We’re shifting from blanket refusals in sensitive areas to a more precise approach focused on preventing real-world harm.”
Jang also said, “The goal is to embrace humility: recognizing how much we don’t know, and positioning ourselves to adapt as we learn.”
Public Figures Are No Longer Treated The Same Way
One major change concerns public figures. Under the updated policy, ChatGPT can generate and modify images of Donald Trump, Elon Musk, and other public figures that OpenAI did not previously allow.
Jang's explanation is that OpenAI does not want to decide who qualifies for image generation and who does not. Instead of acting as the arbiter of status, the company is giving users an opt-out option if they do not want ChatGPT depicting them.
That is a meaningful shift in how the system treats fame, identity, and consent. The previous approach relied more heavily on categorical refusal. The new approach allows more requests while preserving some boundaries around harmful uses.
OpenAI has also said this fits into a wider plan to effectively “uncensor” ChatGPT. In February, the company announced changes to how it trains AI models, with the goal of letting ChatGPT handle more requests, provide diverse perspectives, and reduce the number of topics the chatbot refuses to engage with.
Hateful Symbols And Physical Traits Get A Narrower Test
OpenAI is also changing how ChatGPT handles hateful symbols. In a white paper released Tuesday, the company said it will allow users to “generate hateful symbols,” such as swastikas, in educational or neutral contexts.
The stated boundary is that the request must not “clearly praise or endorse extremist agendas.” That means the same symbol may be treated differently depending on how the user frames the image request.
The company is also changing how it defines “offensive” content. According to Jang, ChatGPT previously refused requests involving physical characteristics, including prompts such as “make this person’s eyes look more Asian” or “make this person heavier.”
TechCrunch reported that, in its testing, ChatGPT's new image generator fulfilled these types of requests. That points to a more context-based moderation system, where the tool may comply with edits that were once rejected outright.
These changes do not mean every sensitive request is accepted. The source article notes that GPT-4o's native image generator still refuses many sensitive queries. It also says the tool has more safeguards around generating images of children than DALL-E 3, ChatGPT's previous AI image generator, according to GPT-4o's white paper.
Style Imitation Keeps One Important Limit
ChatGPT can now mimic the styles of creative studios, including Pixar or Studio Ghibli. That capability helped drive the wave of viral Studio Ghibli memes after the new image generator launched.
OpenAI still restricts imitation of individual living artists' styles. The distinction matters because studio-like visual references and living artist imitation raise different moderation and rights questions inside the product.
The source article notes that this could revive an existing debate about the fair use of copyrighted works in AI training datasets. That debate is not settled by the new image generator, but the product's popularity puts the issue back in view.
For users, the practical result is that ChatGPT image creation now has a wider creative range. For OpenAI, the same expansion creates more pressure to explain where the boundaries sit and why some requests pass while others fail.
The Moderation Debate Is Getting Sharper
OpenAI's policy shift arrives after years of conservative complaints about alleged AI “censorship” by Silicon Valley companies. The source article also points to backlash Google faced after Gemini's AI image generator created multiracial images for queries such as “U.S. founding fathers” and “German soldiers in WWII,” which were obviously inaccurate.
The political scrutiny has continued. Earlier this month, Republican Congressman Jim Jordan sent questions to OpenAI, Google, and other tech giants about potential collusion with the Biden administration to censor AI-generated content.
OpenAI previously told TechCrunch that its moderation changes were not politically motivated. The company said the shift reflects a “long-held belief in giving users more control,” and that its technology is now capable enough to navigate sensitive subjects.
Other major platforms are moving in a similar direction. The source article notes that Silicon Valley giants like Meta and X have adopted similar policies allowing more controversial topics on their platforms.
For now, OpenAI's new image generator is best known for Studio Ghibli-style posts. But the deeper story is about how ChatGPT will handle difficult image requests as it becomes more capable. More control for users may make the product feel less restrictive, but it also gives OpenAI a harder moderation problem to manage.