During a livestream on Tuesday, OpenAI CEO Sam Altman announced the first major update to ChatGPT’s image generation capabilities in over a year.
ChatGPT can now use OpenAI’s GPT-4o model to create and modify images and photos. GPT-4o has long been at the heart of the AI chatbot platform, but until now, the model could only generate and edit text, not images.
According to Altman, native GPT-4o image generation is already available in ChatGPT and Sora, OpenAI’s AI video creation product, for subscribers to the company’s $200/month Pro plan. OpenAI says that the feature will soon be available to ChatGPT users with a Plus subscription and for free, as well as to developers using the company’s API service.
The GPT-4o with image output “thinks” a little longer than the image generation model it actually replaces, the DALL-E 3, to produce what OpenAI describes as more accurate and detailed images. The GPT-4o can edit existing images, including images with people, by transforming them or “painting in” details such as foreground and background objects.
To power the new image manipulation feature, OpenAI told the Wall Street Journal that it trained GPT-4o on “publicly available data” as well as proprietary data from partnerships with companies such as Shutterstock.
Many generative AI vendors view training data as a competitive advantage, so they keep it and any information related to it close to the vest. But training data is also a potential source of intellectual property lawsuits, which is another deterrent for companies that don’t want to disclose too much information.
“We respect the rights of artists in how we create results, and we have a policy that does not allow us to generate images that directly mimic the work of living artists,” said Brad Lightcap, OpenAI’s COO, in a statement to the magazine.
OpenAI offers an opt-out form that allows authors to request that their work be removed from the training datasets. The company also states that it respects requests to prevent its bots from collecting training data, including images, from websites.
ChatGPT’s updated image generation feature comes after Google launched an experimental custom image output for Gemini 2.0 Flash, one of the company’s flagship models. This powerful feature went viral on social media – but not necessarily for the best reasons. The image component of Gemini 2.0 Flash was not well protected, allowing people to remove watermarks and create images with images of copyrighted characters.
