OpenAI releases ChatGPT Images 2.0 with accurate text rendering and brand-style matching

OpenAI launched ChatGPT Images 2.0 this week, marking a shift from what the company calls "decorative" images to production-ready graphics with detailed text rendering and brand-style matching capabilities.

The update is available across all ChatGPT tiers, with more capable language features limited to paid subscriptions using the Thinking model.

Key capabilities tested

Hands-on testing by ZDNET's David Gewirtz revealed several notable improvements:

Text accuracy: The model correctly rendered all text in sketchnotes of the US Bill of Rights, including angled text, with no duplicates or omissions. This contrasts with Google's Nano Banana, which required multiple iterations to achieve text accuracy.

Brand consistency: When provided with a standalone ZDNET logo, Images 2.0 successfully matched brand colors and styling across multiple graphic types, including infographics and sketchnotes. The model maintained consistent brand elements when generating graphics on both white and black backgrounds.

Content interpretation: The system can read article content and convert it into visual formats like infographics, adding contextual information such as pricing data not present in the original text.

Documented errors

Testing revealed systematic issues requiring human oversight:

• Factual inconsistencies: An infographic claimed to feature "9 of the best AI website builders" while displaying only five, the actual number reviewed
• Substitution errors: The model replaced 10Web (a reviewed service) with Durable (an unreviewed competitor) without explanation
• Fabricated data: Generated star ratings for categories not present in source material, with ratings contradicting review text
• Visual artifacts: Minor rendering issues like unexplained lines near logos

When re-prompted with corrections, the model successfully updated the infographic with appropriate information.

Technical context

ZDNet blocks OpenAI from scraping its pages. Ziff Davis, ZDNET's parent company, filed an April 2025 lawsuit against OpenAI alleging copyright infringement in training and operating its AI systems. Testing was conducted using Chrome extension screenshots rather than direct page access.

What this means

ChatGPT Images 2.0 represents a significant technical advance in AI-generated graphics, particularly for text rendering—a persistent weakness in previous image generation models. The ability to maintain brand consistency and interpret complex content into visual formats provides genuine business utility.

However, the documented errors mirror mistakes a junior human designer might produce: incorrect counts, unsolicited substitutions, and fabricated supporting data. The model's tendency to add information beyond source material, while sometimes useful, introduces verification overhead that limits autonomous deployment.

The update positions ChatGPT as viable for first-draft graphics work requiring human review, rather than fully automated visual content generation. Organizations will need to weigh time savings from rapid iteration against quality assurance costs.