OpenAI Releases ChatGPT Images 2.0: AI Model That Excels At Generating Text From Images
Credit: OpenAI
OpenAI has released an updated image generation model, Images 2.0, built into ChatGPT.
The new version is focused on more accurate following of user instructions, improved preservation of fine details, and correct handling of text and icons.
As early as 2024, diffusion AI models were systematically distorting text. According to Asmelash Teka Hadgu, founder and CEO of Lesan AI, the models reconstruct images from noise and learn patterns that cover the majority of pixels, with text occupying a tiny fraction of the area.
Since then, researchers have tried alternative approaches—notably autoregressive models, which predict image content and operate in a manner similar to large language models (LLMs).
OpenAI didn’t reveal the underlying architecture of Images 2.0. The company only explained that the new model can “reason”—search for information online, generate multiple images for a single query, and double-check the results. This allows Images 2.0 to create marketing materials in various sizes and even comics. The AI model also has improved performance with non-Latin scripts, such as Japanese, Korean, Hindi, and Bengali. However, Images 2.0’s knowledge is limited to December 2025, which may impact the accuracy of its generation for queries about recent events.
The introduction of full-fledged text processing transforms the tool from a simple drawing tool into a fully-fledged assistant for creating layouts and presentations. The ability to combine visualization with up-to-date web search significantly simplifies working with data for those who need to quickly prepare a visual report or banner.
