OpenAI added the generation and editing of images to ChatGPT

Goodbye, DALL-E — ChatGPT has learnt to generate and edit images independently
The ChatGPT app on a smartphone screen. Photo: Unsplash

OpenAI CEO Sam Altman has announced the first major update to ChatGPT's image creation capabilities in over a year. Now, AI can not only generate and edit text but also work with photos directly in GPT-4o.

TechCrunch writes about it.

Advertisement

How ChatGPT functionality has expanded

Previously, GPT-4o supported the popular chatbot, but it was used exclusively for text tasks. The new version allows generating and editing images directly in ChatGPT. According to Altman, this feature is already available to users of the USD 200 per month Pro subscription and will soon be available to owners of the Plus plan, the free version of ChatGPT, and developers connecting via the API.

The GPT-4o may take a little longer to process images than the DALL-E 3 model it actually replaces. However, the developers claim that this allows the new system to provide more detailed and accurate results. It is also capable of editing existing photos, including those with people in them, by changing the background or adding new elements.

According to OpenAI, to enable GPT-4o to recognise and generate images, the company used "publicly available information" and its own specialised data obtained through collaboration with partners such as Shutterstock.

For many developers of generative AI, training data is a competitive advantage, so it is often kept secret. At the same time, information about the origin of training data can lead to possible lawsuits, so companies are reluctant to disclose details.

Brad Lightcap, OpenAI's chief operating officer, emphasized that the company respects the rights of artists and does not allow its models to create works that exactly copy the style of contemporary artists. In addition, there is an opt-out form for authors to remove their own work from the training datasets. If a website does not want its content to be crawled by OpenAI bots, the company also complies with such requirements.

As a reminder, OpenAI promised to update the Advanced Voice feature in ChatGPT, which provides real-time communication with AI. The company plans to make it more friendly and get rid of the habit of interrupting users.

We also wrote that OpenAI admitted in recent tests that their GPT-4.5 model is constantly "hallucinating". In more than a third of cases, AI invents facts to disguise them as real.

neural network AI chat bot ChatGPT OpenAI
Advertisement
Advertisement
Advertisement
Advertisement