A tool that eats other tools

OpenAI has launched ChatGPT Images 2.0, an upgraded image model now rolling out across ChatGPT, the API, and the Codex app for Mac. 

The company is positioning it as more than a typical image generator, with features aimed at more practical creative and technical use cases.

One of the biggest changes is better text rendering. 

The model is now more accurate with small text, icons, UI elements, and other fine details, which makes it more useful for mockups, diagrams, and educational visuals. 

OpenAI also says it has improved support for non-Latin languages, including Japanese, Korean, Chinese, Hindi, and Bengali.

Images 2.0 also supports more aspect ratios, from wide banners to tall vertical formats, making it easier to create visuals for slides, mobile, and other layouts. 

OpenAI says it has also improved the model’s realism and style accuracy across formats like photography, cinematic images, manga, and pixel art.

The main new addition is reasoning. In thinking and pro modes, the model can use web context, turn uploaded material into visual explainers, and check its own work for accuracy. 

It can also generate up to eight images from one prompt, including image sets with consistent characters or objects.

In Codex for Mac, the model is now built into the app, giving developers a way to create and compare design ideas without leaving their workspace. 

The model is also available through the API under gpt-image-2, with support for outputs up to 2K resolution.

OpenAI says the model still has some limits. It can struggle with tasks that need precise physical logic, like origami or certain puzzles, and very detailed diagrams may still need manual review. 

The update is rolling out now to ChatGPT and Codex users, while advanced reasoning features are limited to Plus, Pro, and Business plans.

Did they just solve the text-in-image problem? - MV