OpenAI Rolls Out Major GPT-4o Update: Image Generation and Editing Added to ChatGPT 146

Openai officially announced on March 25, 2025 that image editing and visual generation capabilities have been integrated into the GPT-4o model, allowing users to generate images, modify existing visual content, and even design complex materials such as charts, menus, and maps through conversational interaction.

Feature Highlights:

• Real-time iteration: Users can iterate on image requests in real time—for example, asking to generate a "snail in the city" and then refining the scene by changing the background or adding accessories.

• Complex instruction handling: The system is capable of handling more complex image composition instructions.

• Improved text rendering: GPT-4o has made significant improvements in rendering clear and structured text within images, enabling it to better generate professional visual content such as infographics, charts, and logos.

• "Inpainting" feature: Users can "inpaint" existing images, editing foreground and background elements, even when people are present in the photo.

Performance and Speed

Although GPT-4o processes image requests more slowly than its predecessor DALL·E 3, the generated images are more accurate and detailed. Image generation may take up to one minute.

Training Data

OpenAI stated that the image capabilities of GPT-4o were trained using "publicly available data" as well as proprietary content from partners (such as Shutterstock).

Technical Improvements

• Multimodal generation: GPT-4o's image generator is integrated within the same model as its text generator, enabling better combination of text and images.

• Autoregressive generation approach: Unlike diffusion models such as DALL-E 3, GPT-4o uses an autoregressive approach, generating images step by step from left to right and top to bottom, which may be why its text rendering and object binding capabilities are stronger.

This feature update is now available to subscribers of OpenAI's $200-per-month Pro plan, and will then be expanded to free and Plus tier users, as well as developers through the OpenAI API.

Details:https://openai.com/index/introducing-4o-image-generation/