OpenAI’s CEO Sam Altman has announced a significant upgrade to ChatGPT’s image generation capabilities during a livestream.
This update brings the ability for ChatGPT to generate and modify images directly using the GPT-4o model, a feature that was not previously available in the platform.
Until now, GPT-4o was only used for text-based tasks, but it can now handle the creation and editing of images and photos.
The new image feature is accessible today to subscribers of OpenAI’s $200-per-month Pro plan, with plans to roll it out soon to Plus users and free users of ChatGPT, as well as developers using OpenAI’s API service.
The upgraded GPT-4o model takes slightly longer to generate images compared to its predecessor, DALL-E 3, but it produces more detailed and accurate results.
The model can edit existing images, including those featuring people, by transforming them or adding details to foreground and background elements.
OpenAI trained GPT-4o with publicly available data and proprietary data through partnerships with companies such as Shutterstock.
This move follows concerns in the AI community about how generative models are trained using large datasets, and OpenAI ensures it respects artist rights by having policies in place that prevent mimicking living artists’ works.
OpenAI has also introduced an opt-out form for creators to request removal of their works from the training datasets, and the company honors requests to stop its web-scraping bots from collecting data, including images, from websites.
This upgrade from OpenAI comes shortly after Google introduced a similar image-generation feature in its Gemini 2.0 Flash model, which gained attention on social media, though not without controversy.
