OpenAI ships image model with better text
- OpenAI on April 21 introduced ChatGPT Images 2.0 and gpt-image-2, a new image model aimed at clearer text, tighter layouts, and stronger edits. - OpenAI said the model improves multilingual text rendering and visual reasoning, and made gpt-image-2 available the same day in the API and Codex. - The release extends OpenAI’s shift from DALL·E-era art tools toward production graphics and editing workflows. (openai.com)
OpenAI on April 21 released ChatGPT Images 2.0 and a matching API model, gpt-image-2, with a pitch centered on cleaner text inside images. (openai.com) (community.openai.com) Image generators have long been good at posters and product shots but bad at letters, often turning words into gibberish. OpenAI said the new model improves text rendering, multilingual output, and visual reasoning. (openai.com) In OpenAI’s developer documentation, image generation now lists gpt-image-2 as the latest model, alongside older gpt-image-1.5, gpt-image-1, and gpt-image-1-mini options. The company says developers can use it through the Image API or the Responses API. (developers.openai.com 1) (developers.openai.com 2) That matters because text is what turns a synthetic image into a usable flyer, menu, mock app screen, label, or ad. OpenAI’s product post frames the upgrade as a move toward “workhorse” graphics that carry precise information, not just visual style. (openai.com 1) (openai.com 2) The company has been pushing in that direction for more than a year. In its earlier 4o image-generation launch, OpenAI said the goal was better prompt following and more accurate text for logos, diagrams, and other information-heavy images. (openai.com) The API pages also show how OpenAI is reorganizing its image lineup. GPT Image models now sit at the center of the platform, while gpt-image-1 is labeled a previous model and newer guides point developers to gpt-image-2. (developers.openai.com 1) (developers.openai.com 2) OpenAI did not present a public benchmark in the launch materials showing exact text-accuracy gains over the prior model. It instead emphasized example outputs, stronger editing, layout control, and availability inside ChatGPT, the API, and Codex. (openai.com) (community.openai.com) That leaves the practical test to users building with it: whether a generated package mockup spells every word correctly, and whether an interface comp keeps the button labels where they belong. (openai.com) (developers.openai.com)
