• “Thinking” capability links web search with scene analysis before image generation
• Improved consistency enables producing a matched set of images from a single prompt

OpenAI has announced the launch of a new update called “ChatGPT Images 2.0”, a major upgrade in image generation capabilities built on the “GPT Image 2” model.

The system introduces what the company describes as “thinking capabilities”, allowing it to develop a deeper understanding of context before generating images

The new update relies on a system that can search the internet for supporting information, analyze the structure of the requested scene, and then carry out image generation with greater accuracy and consistency. It can also produce up to eight interconnected images from a single prompt, while preserving the same elements, characters, and visual style.

The system also delivers notable improvements in text rendering inside images, a long-standing challenge in earlier AI models. In addition, it offers stronger adherence to user instructions, supports resolutions up to 2K, and provides a wider range of aspect ratios, including horizontal, square, and vertical formats.

The company said the “thinking” feature will be available to Plus, Pro, Business, and Enterprise subscribers, while all users will benefit from baseline improvements in image quality and precision.

This update represents a shift away from traditional one-step generation toward a system that combines search, planning, and verification before producing an image, opening the door to more advanced applications in design, advertising, and visual content production.

The new system is expected to accelerate the creation of consistent visual content, especially in multi-scene projects such as comics or marketing campaigns, while reducing the need for complex manual editing.

However, experts warn that increased realism and improved in-image text may make verification and content monitoring more important, particularly as such capabilities could also be misused to generate highly convincing visual content.