OpenAI Eases ChatGPT's Image Creation Restrictions

# ChatGPT # openai

This week, OpenAI unveiled a new image generator for ChatGPT that quickly gained popularity due to its ability to produce Studio Ghibli-inspired visuals. But beyond the charming illustrations, the introduction of GPT-4o’s native image generator marks a significant enhancement in ChatGPT's capabilities, including better picture editing, text rendering, and spatial representation. However, one of the most striking changes OpenAI implemented this week pertains to its content moderation policies. These updated policies now permit ChatGPT to generate images of public figures, hateful symbols, and racial features upon request, a shift from its previous stance of outright rejection. According to a blog post by OpenAI’s model behavior lead, Joanne Jang, published on Thursday, the company has "evolved" its approach. Jang explained, “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.” These changes appear to align with OpenAI’s broader strategy to "uncensor" ChatGPT. In February, the company announced plans to modify its AI model training methods, aiming to enable ChatGPT to handle more requests, provide diverse perspectives, and minimize the topics it refuses to engage with. Under the new policy, ChatGPT can now create and alter images of public figures such as Donald Trump and Elon Musk, which were previously off-limits. Jang emphasized that OpenAI doesn't want to dictate who should or shouldn't be depicted by ChatGPT, offering an opt-out option for those who prefer not to be included. In a white paper released on Tuesday, OpenAI also stated that it will allow users to generate images of hateful symbols, like swastikas, in educational or neutral contexts, provided they don’t promote extremist agendas. Furthermore, OpenAI is redefining what constitutes "offensive" content. Previously, ChatGPT would reject prompts about physical characteristics, such as requests to "make this person’s eyes look more Asian" or "make this person heavier." Our testing at TechCrunch revealed that ChatGPT’s new image generator now complies with these types of requests. Additionally, while ChatGPT can now mimic the styles of creative studios like Pixar or Studio Ghibli, it still refrains from imitating the styles of living artists, potentially reigniting debates over the use of copyrighted material in AI training data. It’s important to note that OpenAI isn’t completely removing all safeguards. GPT-4o’s native image generator still declines many sensitive queries and has more stringent protections around generating images of children than its predecessor, DALL-E 3, as outlined in GPT-4o’s white paper. This relaxation of guardrails comes after years of criticism from conservative circles about alleged AI "censorship" by Silicon Valley firms. Google, for instance, faced backlash for its Gemini AI image generator, which produced historically inaccurate multiracial images for queries like "U.S. founding fathers" and "German soldiers in WWII." The ongoing culture war surrounding AI content moderation may be reaching a critical point. Earlier this month, Republican Congressman Jim Jordan questioned OpenAI, Google, and other tech companies about possible collusion with the Biden administration to censor AI-generated content. In a previous statement to TechCrunch, OpenAI denied that its content moderation changes were politically driven. Instead, the company maintains that these shifts reflect a "long-held belief in giving users more control," and its technology is now advanced enough to handle sensitive topics responsibly. Regardless of the motivations behind these changes, the timing seems opportune, considering the potential for regulatory scrutiny under the Trump administration. Other tech giants like Meta and X have also adopted similar policies, allowing more controversial topics on their platforms. While OpenAI’s new image generator has so far only sparked some viral Studio Ghibli memes, the long-term implications of these policy shifts remain uncertain. ChatGPT’s recent changes might align well with the Trump administration's views, but allowing an AI chatbot to tackle sensitive questions could soon put OpenAI in a challenging position. > OpenAI’s viral Studio Ghibli moment highlights AI copyright concerns > ChatGPT’s image-generation feature gets an upgrade

Related article

Comments (35)