OpenAI Beefs Up ChatGPT’s Image Generation Model

OpenAI has released an upgraded version of its image generation engine called ChatGPT Images 2.0. This new model enhances the quality and accuracy of AI-generated images, allowing for the creation of more detailed visuals and better-rendered text—especially in English. It also introduces the ability to produce multiple images from a single prompt, letting users design things like a booklet or collage in one go.

Key Improvements in Images 2.0

• More Images per Prompt: Users can now generate sets of images or documents at once, instead of one picture at a time.
• Enhanced Text Rendering: The model does a much better job at accurately rendering text in images, with significantly fewer errors in English words and characters.
• Recent Knowledge Cutoff: The knowledge base powering Images 2.0 is updated as of December 2025, so the rendered content is more up-to-date.
• Customizable Aspect Ratios: Users can select image dimensions, ranging from wide panoramas (3:1) to tall portraits (1:3), for greater creative control.

Performance Observations

Testing reveals that ChatGPT Images 2.0 excels at producing images with English text, reducing the common mishaps and distorted lettering that plagued earlier versions. However, attempts to use non-English languages, such as Chinese, often result in jumbled or nonsensical text. While the model can simulate the look of non-English scripts, accurate translation and contextual use remain challenges.

The introduction of this model follows a trend where new AI image tools often spark viral trends and increased user engagement. As with other recent AI advancements, ongoing use and feedback—especially from users who operate in languages other than English—could help the technology expand its capabilities and accuracy over time.

Looking Ahead

ChatGPT Images 2.0 represents a significant step forward for AI-generated art, particularly for English speakers. Although outputs in other languages have room for improvement, OpenAI is likely to iterate further based on user data. The model is available globally for all users, with additional features accessible to paying subscribers.

For more information, you can read the original article by Reece Rogers at WIRED.