ChatGPT Image Generation Upgrade: What's New?

OpenAI Enhances ChatGPT with Advanced Image Generation

A significant upgrade to ChatGPT’s image creation abilities was unveiled by OpenAI CEO Sam Altman during a livestream event on Tuesday. This marks the first substantial improvement to these capabilities in over a year.

ChatGPT now utilizes the GPT-4o model to natively generate and modify images and photographs. While GPT-4o has been the foundation of the AI chatbot platform for some time, its functionality was previously limited to text-based generation and editing.

Availability of the New Feature

The native image generation feature powered by GPT-4o is currently available within both ChatGPT and Sora, OpenAI’s AI video generation tool. Access is granted to subscribers of the $200 per month Pro plan.

OpenAI has announced that the feature will soon be extended to Plus and free users of ChatGPT, as well as developers who leverage the company’s API service.

Improved Image Quality and Editing Capabilities

Compared to the previously used image-generation model, DALL-E 3, GPT-4o takes a slightly longer processing time to produce images. This is intended to result in outputs that are more accurate and detailed.

GPT-4o also introduces the ability to edit existing images, including those containing people. This allows for transformations and the addition or alteration of details within the foreground and background.

Data Sources for Training

According to information shared with the Wall Street Journal, OpenAI trained GPT-4o on a combination of “publicly available data” and proprietary data obtained through partnerships, such as the one with Shutterstock.

The data used for training generative AI models is often considered a key competitive advantage, leading many companies to maintain strict control over it.

Addressing Artist Rights and Copyright Concerns

OpenAI emphasizes its commitment to respecting artists’ rights. Policies are in place to prevent the generation of images that directly replicate the style of living artists.

Brad Lightcap, OpenAI’s chief operating officer, stated this commitment in a statement to the Journal.

Creators have the option to request the removal of their work from OpenAI’s training datasets through an opt-out form.

The company also respects requests to prevent its web-scraping bots from collecting training data, including images, from specific websites.

Comparison to Google’s Gemini

ChatGPT’s upgraded image generation follows Google’s release of an experimental native image output feature for Gemini 2.0 Flash. This feature gained significant attention on social media.

However, Gemini 2.0 Flash’s image component faced criticism due to a lack of sufficient safeguards, which allowed for the removal of watermarks and the creation of images featuring copyrighted characters.

Update Information

This article has been updated to include OpenAI’s statement to the Wall Street Journal regarding the training data used for GPT-4o, as of 12pm PT.