OpenAI upgrades ChatGPT with image generation and editing powered by GPT-4o

During a livestream on Tuesday, OpenAI CEO Sam Altman announced a significant upgrade to ChatGPT’s image-generation capabilities.

Until now, the image generation and editing features powered by the DALL-E model have been available within the chat interface and through the image generation app.

Now, it’s powered by the GPT-4o model, which has only generated and edited text until now.

The new feature, available today for subscribers of OpenAI’s $200-a-month Pro plan, enables ChatGPT to create and modify images. That includes editing existing images with people, through a process called “inpainting”.

OpenAI says it trained GPT-4o on publicly available and proprietary data. That includes partnerships with companies like Shutterstock. However, it has ensured that artists’ rights are respected. It has policies in place to prevent the generation of images that mimic the work of living artists.

OpenAI has also set up an opt-out form for creators who wish to have their works removed from the training datasets.

This is very much in line with Google’s introduction of native image output within its Gemini 2.0 Flash model. That launch was heavily criticised for the lack of guardrails.

Altman’s goal seems to be making image generation more useful. He pointed out that it could be used to create diagrams and infographics, as well as social media posts.

He reiterated the commitment to blocking content that violates its policies, such as child sexual abuse materials.

All generated images will carry C2PA metadata to ensure provenance can be tracked. The feature will also be available to Enterprise and Edu users soon.