ChatGPT Images 2.0 Shows Major Leap in AI Image Generation With Strong Text Rendering Capabilities

Artificial intelligence image generation has taken a noticeable step forward with the launch of ChatGPT Images 2.0, a new model from OpenAI that is significantly better at generating readable and accurate text within images. Just a few years ago, AI-generated visuals often contained distorted or nonsensical words, making them easy to identify. Today, that gap is rapidly closing.

Early AI image tools frequently struggled with spelling and layout, especially when generating structured content like menus or posters. Users often saw garbled text or fictional words that broke realism. However, with Images 2.0, outputs are now detailed enough to resemble real-world designs, including restaurant menus that could plausibly be used in actual businesses.

The improvement is largely linked to advancements in how AI models generate images. Older systems relied heavily on diffusion models, which reconstruct visuals from noise and often mishandle fine details like letters. Newer approaches, including more LLM-like methods, allow models to better predict structure and maintain coherence across complex designs, improving text accuracy and layout consistency.

OpenAI has described Images 2.0 as having “thinking capabilities,” enabling it to refine prompts, generate multiple image variations, and even check its own outputs. It can also produce marketing materials in different formats, including multi-panel comic strips, and better handle non-Latin scripts such as Japanese, Korean, Hindi, and Bengali. However, the company has not fully disclosed the exact model architecture behind it.

The rollout of Images 2.0 will be available to all ChatGPT and Codex users, with expanded features for paid subscribers. Developers will also gain access through the gpt-image-2 API, with pricing based on output quality and resolution, marking another step toward more practical, production-ready AI-generated visual content.

source: Techcrunch