During a livestream on Tuesday, OpenAI CEO Sam Altman announced the first significant enhancement to ChatGPT’s image-generation abilities in over a year.
ChatGPT can now utilize the GPT-4o model to create and modify images and photos natively. While GPT-4o has powered the AI chatbot platform for some time, it previously only generated and edited text.
Altman revealed that the native image generation using GPT-4o is now available in ChatGPT and Sora, OpenAI’s AI video-generation tool, for subscribers of the $200-a-month Pro plan. The feature will soon be accessible to Plus and free ChatGPT users, as well as developers using the company’s API.
GPT-4o takes a bit longer to generate images compared to the model it replaces, DALL-E 3, resulting in what OpenAI describes as more accurate and detailed visuals. It can also edit existing images, including those featuring people, by transforming them or “inpainting” details such as foreground and background objects.
To enable this new image feature, OpenAI trained GPT-4o on publicly available data and proprietary data from partnerships with companies like Shutterstock.
Many generative AI companies consider their training data a competitive edge, often keeping it confidential to avoid potential IP lawsuits.
ICYMT: Geyse Confirms Manchester United Exit
“We respect artists’ rights in how we produce outputs and have policies to prevent us from generating images that closely mimic any living artist’s work,” stated Brad Lightcap, OpenAI’s COO.
OpenAI provides an opt-out form for creators wishing to remove their works from training datasets and honors requests to prevent its web-scraping bots from collecting data from websites.
ChatGPT’s enhanced image-generation feature comes shortly after Google’s experimental image output for Gemini 2.0 Flash gained attention on social media, albeit for controversial reasons due to its lack of guardrails.
This article was updated at 12 PM PT to include OpenAI’s statement regarding GPT-4o’s training data.
SOURCE: TECH CRUNCH