OpenAI has introduced ChatGPT Images 2.0, an upgraded image generation engine that shifts its focus from basic visuals to more functional, text-rich graphics aimed at real-world use. Earlier this week, OpenAI officially unveiled the new system, highlighting its ability to move beyond simple decorative images into producing full-page visuals with detailed and accurate text. The reviewer, who had early access to a pre-release version, noted that while the earlier build showed promise, it struggled with consistent brand reproduction, particularly when handling specific logos. With the public release now available across all ChatGPT tiers, further testing was conducted using a ChatGPT Plus account with advanced reasoning features enabled.
One of the primary tests focused on brand consistency using ZDNET visuals. Instead of relying on existing web pages, a standalone ZDNET logo was uploaded alongside prompts to guide the system. This significantly improved performance, allowing the model to generate infographics that not only preserved the logo accurately but also matched the publication’s color scheme and visual identity. The most notable improvement, however, was in text rendering. The generated images included detailed and correctly spelled text, even in smaller or angled sections, an area where previous AI image tools often struggled. The testing expanded into more complex creative tasks, including generating sketchnotes of the US Bill of Rights. Compared to earlier experiments with competing tools like Google’s Gemini Nano Banana, which required repeated prompting to correct textual inaccuracies, ChatGPT Images 2.0 performed more reliably. The generated sketchnotes contained complete and accurate text without duplication or omissions. Visual styling also aligned with the requested ZDNET branding, though minor layout issues remained, such as less optimal logo placement in certain variations. These were considered fixable through additional prompt refinement.
Despite the improvements, the testing revealed some limitations when converting long-form content into infographics. In one example, the model was asked to visualize an article about AI website builders. While the output was visually structured and included additional contextual information such as pricing, several factual inconsistencies appeared. The infographic incorrectly referenced nine tools instead of five, replaced one reviewed platform with an unrelated competitor, and introduced star ratings that were not part of the original content. These inaccuracies highlighted a tendency for the model to supplement or reinterpret information rather than strictly adhere to source material. Additional minor design issues were also observed, including small visual artifacts that did not align with the rest of the composition. However, when corrections were provided through follow-up prompts, the system was able to adjust the infographic and improve accuracy, indicating that iterative prompting remains an important part of the workflow.
Overall, ChatGPT Images 2.0 represents a noticeable improvement over earlier versions, particularly in its ability to combine visual design with structured and readable text. The tool demonstrates clear potential for professional applications such as content marketing, design prototyping, and branded media creation. At the same time, the presence of subtle factual errors suggests that human oversight is still necessary, especially when accuracy is critical.
Follow the SPIN IDG WhatsApp Channel for updates across the Smart Pakistan Insights Network covering all of Pakistan’s technology ecosystem.
